Max Planck Intelligent Systems Colloquium

Das "Max Planck Intelligent Systems Colloquium" ist eine Vortragsreihe, die während des Semesters an beiden Standorten des Instituts stattfindet.

Weltweit anerkannte Wissenschaftler sprechen über Themen, welche von breitem Interesse sind, für die an intelligenten Systemen interessierte Wissenschaftsgemeinde.

Monat:

Max Planck Intelligent Systems Colloquium, Stuttgart

11311 1508408014

Wearable haptics for VR and AR

[mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

A General Overview of Space Robotics

[mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Max Planck Intelligent Systems Colloquium, Stuttgart

Max Planck Intelligent Systems Colloquium, Stuttgart

8864 1485856382

Minimalistic approach in the design and modeling of bio-inspired robots

Miniature crawling robots, have a variety of applications including medical procedures, search and rescue operations, maintenance, reconnaissance, environmental monitoring, assisted agriculture, cave exploration, mapping, studying biological organisms, security, defense, exploration of hazardous environments. Their small size allows them to navigate in remote areas otherwise inaccessible to wheeled or larger robots such as collapsed building and caves and biological vessels, while their low cost makes them disposable and allows their use in large quantities and swarm formations. As miniature crawling robots at this scale are minimally actuated or under-actuated, the main challenges are in achieving stability at high velocities over varying terrain, reducing the cost of transport, overcoming obstacles, controlling jumping and landing, adhering quickly to different surfaces, transitioning between horizontal to vertical motion, and climbing. This talk will address some of the modeling and actuation challenges of crawling robots inside biological vessels and outdoors, while taking into account contact compliance and sliding. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

8863 1485856306

Design of mechanical properties of soft elastic materials through network architecture

Soft elastic materials are usually made from networks of flexible polymer chains, where entropic elasticity controls the storage and release of elastic energy. They can, in principle, deform reversibly up to the point where polymer chains are highly oriented in the tensile direction, but in practice these elastic networks fail often early due to crack propagation. This is particular true for very soft ones such as gels and for simple networks of elastomers. In recent years new reversibly elastic materials have been developed with a non-random network architecture specifically designed to delay fracture. Costantino Creton is going to discuss and rationalize these strategies and show an example of how sacrificial bonds inside the material can provide significant toughening in the fracture mechanics sense, while retaining the reversible elasticity of the material. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

7864 1480933606

Neuromodulation: a fast growing area of Clinical Neuroscience

The main topic of investigation in Yasin Temel's group is to develop new ways of modulating deep brain activity to treat severe neurological and psychiatric disorders. Different lines of neuroscience research show that altered activity in populations of neurons are responsible for key symptoms, such as hypokinesia in Parkinson's disease, anhedonia in Depression and motor tics in Tourette syndrome. In Prof. Temel’s research, a close collaboration exists between clinicians, neurobiologists and engineers, using a wide range of models, from in vitro set-ups to patients. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Theoretical approaches for problems in biolocomotion

Fluid mechanics plays a crucial role in many cellular processes. One example is the external fluid mechanics of motile cells such as bacteria, spermatozoa, and essentially half of the microorganisms on earth. These organisms typically possess flagella, slender whiplike appendages which are actuated in a periodic fashion in a fluid environment, thereby giving rise to propulsion. In this talk, I will review, along with classical work in the field, research in my group over the last few years attempting to develop and use different theoretical methods to capture these biological flows.  [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Novel Alliance between Advanced Materials and Volatile Biomarkers for Non-Invasive Medical Evaluation

Medical evaluation for early detection of a disease is required to reveal groups of individuals from the general population in whom the likelihood of the disease is increased and who could benefit from further medical evaluation. The ideal medical evaluation is high-accuracy, low-cost, non-invasive, easily repeatable, effortlessly operated by a lay-person and has minimal impact on the person’s daily activities. In our research, we tackle these requirements by the development of novel solid-state and self-healable flexible1 devices/sensors that are based on advanced functional materials as well as electronic sensory nanoarrays for profiling volatile biomarkers that are emitted from cells in the affected area and that can be detected either via the exhaled breath or from the skin, without going invasively into the human body. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Integrative Design Computation and Robotic Fabrication in Architecture

A new understanding of the material in architecture is fast emerging. Computational designers are no longer conceiving of the digital realm as separate from the physical world. Instead computation is being regarded as the key interface for material exploration and vice versa. In this lecture, Achim Menges will discuss how this represents a significant perceptual shift in which the materiality of architecture is no longer seen to be a fixed property and passive receptor of form, but is transformed into an active generator of design and an adaptive agent of architectural performance. In stark contrast to previous linear and mechanistic modes of fabrication and construction, materialisation is now beginning to coexist with design as explorative robotic processes. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

6644 1474538394

Soft, Stretchable, and Reconfigurable Materials for Electronics and Actuators?

This talk will describe efforts in our research group to control the shape and function of soft materials (liquid metals, polymers and hydrogels) for applications that include stretchable electronics, soft robots, and self-folding polymer sheets.  The research harnesses interfacial phenomena, micro fabrication, patterning, and thin films.  [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Biologically inspired model-free robotic design optimization: any hope?

There are many physical phenomena in the real world that cannot be mathematically modeled. If we consider machine design problems that cope with such physical dynamics, we need heuristic search of design in physical systems, which we call the model-free (robotic) design optimization. We are exploring how such a new paradigm of design optimization processes can be realized, though there are a number of known challenges such as the dimensionality problem, the scalability problem, and the reality gap. In this talk, I would like to introduce some of the attempts in our laboratory and discuss challenges and perspectives. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

6236 1469185270

Actuating Magnetic Micro-Scale Robots

Micro-scale mobile robots can physically access small spaces in a versatile and non-invasive manner. Such microrobots under 1 mm in size have potential unique applications for object manipulation, local sensing and cargo delivery in healthcare, microfluidics and advanced materials fabrication. These devices are powered and controlled remotely using externally-applied magnetic fields for motion in 2D and 3D. This talk will introduce our experimental work in a new type of field generation system using an array of permanent magnets, as well as use of magnetic field actuation for multi-agent formation control. Other uses of remote magnetic actuation including in-situ fabrication of microstructures for micromanipulation and microfluidics will be shown. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

6030 1467977701

Multi-modal Endoscopic Imaging Systems for Optical Biopsy

Optical biopsy is a diagnostic method which uses light to obtain immediate histological results without the need to remove tissue. Techniques such as Raman or elastic scattering spectroscopy, fluorescence imaging, confocal microscopy, photo-acoustic imaging, or optical coherence tomography (OCT) have been demonstrated as viable means to distinguish between benign and malignant tissue. Regardless of the target pathology, however, no single optical method is yet to attain the selectivity and specificity of traditional biopsy. This is why multi-modal imaging, with the potential to provide comprehensive tissue characterization through complementary modalities, will be the key to the success of optical biology. Yet, implementing multiple imaging modalities with vastly different optical system requirements within the confines of an endoscope head is a daunting task. Based on our microfabricated micro-optical bench technology, we develop novel endoscopic images probes that can combine different confocal and full-field imaging modalities implemented through separate optical paths, which can be designed and optimized independently. Through their compact footprint and enhanced functionality, these probes provide depth-resolved guiding capability for existing laparoscopes and represent a major step towards a new class of multi-modal endoscopic imaging probes. The talk will focus on the design, implementation and the characterization of a few probe examples and discuss their potential for clinical diagnostic use.  [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

5885 1466497340

Engineering proteins for non-covalent functionalization of single wall carbon nanotubes for targeted subcellular delivery

Single wall carbon nanotubes (SWCNTs) can be non-covalently dispersed with surfactants, polymers and proteins. We have observed that protein coated SWCNTs enter numerous cell types by the millions, more than an order of magnitude more than polymer coated SWCNTs. In addition to increasing uptake, protein coatings can impart function to nanomaterials. While many contemporary studies have focused on using proteins to target SWCNTs and nanomaterials to specific cells, we utilize proteins to target SWCNTs to regions within cells. Here, we will show the utility of modified albumin proteins and engineered native cellular proteins as dispersing agents for SWCNTs. To localize SWCNTs to the nucleus we dispersed SWCNTs with the tail domain of a nucleoskeletal protein lamin B1 (LB1) engineered from the full-length LMNB1 cDNA. The low molecular weight globular protein has a central hydrophobic core for nanotube association and stabilization and an exposed nuclear localization sequence to promote active nuclear import. SWCNTs-LB1 enter HeLa cells and localize to the nucleus of cells; we visualize localization with Raman spectroscopy and NIR fluorescence imaging of SWCNTs and interaction with DNA by fluorescence lifetime imaging microscopy. We have used similar techniques to show cellular uptake of bovine serum albumin (BSA)-coated SWCNTs within cells. We see localization of SWCNTs-BSA within the endosomal and metabolic processing compartments. Similarly to LB1, BSA is a small globular protein, and the hydrophobic cleft of the BSA binds to SWCNTs. BSA has smaller hydrophobic regions, independent of the hydrophobic cleft, that can be loaded with drugs or fluorophores. In vitro, we show that denaturation or enzymatic processing of the BSA releases the small molecules. SWCNTs-BSA in which the BSA are pre-loaded with rhodamine drastically increases small molecule delivery in culture. We have determined spatial and concentration distribution of rhodamine within the cells, and signal is both coincident and distinct from SWCNTs as measured by NIR fluorescence and Raman. We demonstrate efficacy of this approach by delivering a fluorescent chemotherapeutic drug daunomycin that reduces proliferation in cancer cells. Together, our results demonstrates a pathway to increase the delivery of a wide variety of drugs to cells through SWCNTs coated with albumin pre-loaded with drug molecules. The complexity of protein structures allows for multimodal modification and manipulation of cellular processes in addition to SWCNT dispersion. The modification of native cellular proteins as non-covalent dispersing agents to provide specific transport opens new possibilities to utilize both SWCNT and protein properties for multifunctional subcellular targeting applications. Specifically, nuclear targeting will allow delivery of anticancer therapies, genetic treatments, or DNA, which in turn will promote development of novel cellular therapies. Using the transport properties of albumins and intracellular processing by enzymes allows delivery of native molecules and drugs. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

5884 1466495965

Robot, wetting, and micromanipulation

At small scales, the surface force become more dominant than the volume force. In this talk, I will discuss the application of wetting and surface tension to enable dexterous or precise manipulation of micro components, including droplet assisted high-speed and high-precision assembly technique, and their applications, as well as our work in associated wetting study. Additionally, I will also give a brief overview of our other activities in robotic micromanipulation and materials characterization. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

5843 1465834324

Programmable assembly of hybrid colloidal molecules

Patchy colloids with directional interactions have spurred enormous interest in the research community, where they have been applied for the formation of colloidal superstructures or as self-propelled, actively moving objects. They have also been termed ‘colloidal molecules’ because of their compositional symmetries and their directional interaction capabilities. Still, it remained a tremendous challenge to prepare colloidal molecules controlling their geometry, composition and functionality, independently. We developed sequential capillarity-assisted particle assembly (sCAPA) to prepare colloidal molecules of high compositional complexity in a variety of shapes. sCAPA is based on the well-known capillary assembly on topographically patterned templates. Some simple – yet important – modifications of the process allow us to selectively deposit only a single colloidal particle per assembly step [1,2]. As a consequence, it becomes possible to independently define the geometry and the composition of the colloidal molecules: the prior by the shape of the trap and the latter by the sequence of assembly steps. In further steps, the assembled particles can be linked and the colloidal molecules released and dispersed in water [2]. The assembly process works for particles from a wide variety of materials. We prepared chain-like colloidal molecules that resemble barcodes, block copolymer chains and surfactants from polystyrene micro spheres, silica spheres, and silica spheres containing magnetic nanoparticles [2]. The programmability of our approach opens up new directions not only to assemble and study complex materials with single-particle-level control, but also to fabricate new microscale devices for sensing, patterning and delivery applications. [1] S. Ni, J. Leemann, H. Wolf and L. Isa, Faraday Discuss., 2015, 181, 225 [2] S. Ni, J. Leemann, I. Buttinoni, L. Isa and H. Wolf, Science Advances, 2016, 2, e1501779 [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

5689 1464878621

New ideas for designing and implementing robot swarms

Swarm robotics is about constructing and controlling swarms of autonomous robots that cooperate to perform tasks that go beyond the capabilities of the single robots in the swarm. Robot swarms are advertised as robotic systems with inherent fault tolerance, scalability and parallel operation. However, in current robot swarms these properties are present only to a very limited extent. In the talk, I will propose and discuss a few research directions to address and possibly overcome these limitations. In particular, I will present (i) property-driven design, a methodology to address the micro-macro link problem; (ii) design patterns for robot swarms, that provide formal guidelines to deal with recurring problems; and (iii) robots with mergeable nervous systems, which introduces elements of centralisation in the organisation of a robot swarm without losing the benefits of self-organisation. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

5213 1459757101

New Directions in Medical Robotics

My lab creates medical robots not only for minimally invasive surgery, but also for targeted drug delivery and for inducing tissue growth. This talk will describe three of our technologies. The first is a type of continuum robot that is based on concentrically combining pre-curved elastic tubes. We are creating these robots with the goal of increasing the safety and effectiveness of intracardiac and neurosurgical procedures. The second technology consists of tetherless robots that are powered, controlled and imaged using an MRI scanner. These devices vary from patient-mounted needle-driving robots to capsules that can move inside fluid-filled body lumens. The third technology is comprised of robotic implants that apply traction forces to induce soft tissue growth. Applications include lengthening the esophagus and bowel for the treatment of congenital defects and disease. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4336 1456918814

Targeted Drug Delivery Systems: Needle Steering and Medical Microrobotics

The talk provides an overview of two ongoing research topics within the Surgical Robotics Lab in the area of targeted drug delivery systems, i.e., needle steering and medical microrobotics. In the first part of the talk, I will combine needle deflection models with image-guided techniques to steer flexible needles to a moving target. Two different models for predicting needle deflection undergoing multiple bends are presented. The first is a kinematics-based model, and the second model predicts needle deflection based on the mechanics of needle-tissue interaction. The models are validated using double bend experiments in soft-tissue simulants, and also using a needle embedded with Fiber Bragg Grating sensors. The kinematics-based model is used for steering the needles under image-guidance. The proposed steering algorithm is demonstrated using camera and ultrasound images as feedback while compensating for target motion. The algorithm is also used to track a needle undergoing multiple bends in 3D using a 2D ultrasound probe. In the second half of the talk, I will discuss how wirelessly controlled agents might offer advantages in terms of reduced invasiveness and untethered access to deep-­seated regions within the human body. On that account, this talk covers the closed-loop control of microparticles, magnetotactic bacteria, microjets, and magnetosperms. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4512 1456905691

Self-Assembled Peptide Nanostructures for Functional Materials

This lecture will illustrate concepts of making materials, which mimic the structure and function of the biological materials through programmed self-assembly of small molecules and their applications in functional materials. The self-assembly mechanism that forms the supramolecular aggregates involves noncovalent interactions such as hydrogen bonds, electrostatic and hydrophobic interactions. Diverse functional groups were incorporated into nanostructures, for example bioactive peptide sequences and metal chelating groups as well as hydrophobic motifs that include alkyl chains, steroid rings, and aromatic systems. The potential impact of these nanostructures on functional materials will be discussed. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4049 1453212665

Biomimetic adhesive microstructures as an approach to understand functioning of biological systems

Biological hairy attachment systems have robust adhesion and high reliability of contact. Previous comparative experimental studies on biological systems showed the way to development of novel glue-free adhesives. While producing the reversible adhesives, mimicking the gecko attachment system, still remains the main direction of research in the field, very convincing results have been achieved in manufacturing adhesive microstructures inspired by male chrysomelid beetles. Comparative studies on microstructures with different contact geometries showed that beetle-inspired mushroom-shaped adhesive microstructure (MSAMS) even outperform the gecko-inspired spatula-shaped geometry under certain conditions. Adhesion of MSAMS is reversible and even stronger under water. MSAMS demonstrated stick-slip free friction and lower impact of contamination by particles. MSAMS can keep its adhesive capability over thousands of attachment cycles. On rough substrates, their performance can be enhanced by the introduction of fluid into the contact zone. Additionally, the development of MSAMS provides an opportunity for biologists to run experiments, which would be otherwise only hardly possible with real biological system. The present lecture discusses how the knowledge obtained from studies on MSAMS can be applied to understanding function of biological adhesive systems of insects. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4048 1453210922

Colloidal Self-Assembly: From Simple Building Blocks to Functional Materials

The astonishing variety of functionalities found in nature is almost always based on a self-assembly of relatively simple building blocks over multiple length scales. This concept continues to inspire scientists that seek possibilities to create functional materials in a cheap, fast and simple way. Colloidal particles are interesting in that respect as they can be synthesized with high uniformity and precision to yield building blocks with nanoscale dimensions. In the simplest case, such particles are of spherical shape and do not have any special properties by themselves. However, their uniform size allows such spheres to assemble into highly ordered superstructures, very much like oranges or apples on display in a supermarket. In my presentation, I will present examples on how functional, macroscopic properties can emerge from the ordered arrangement of these simple, nanoscale building blocks. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4050 1453213131

Micro-Optics, MEMS, and Applications in Sensing, Imaging, and Displays

Micro-actuators combined with laser light sources and micro-optics enable a number of powerful applications in sensing, imaging, and displays. I’ll describe some of the devices and microsystems developed in our lab, which have been licensed for commercialization: MEMS scanners, pico-projector, 3D and wearable display, MEMS sensor array with integrated optical readout for thermal imaging, and cantilever biosensors for multianalyte detection in a disposable microfluidic cartridge. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

4042 1453123737

Robotic technologies, micro-technologies and targeted therapy: challenges and opportunities

This talk will introduce the key aspects of targeted therapy starting from the speaker experience on robotics for minimally invasive and computer assisted surgery. The quest for miniaturization and natural access to the targeted pathologies led to the development of diagnostic and surgical tools to be delivered with an endoluminal and transluminal approach - such as endoscopic capsules - and to be controlled and propelled by remote operation schemes from outside. In addition to the traditional control of remote devices into the body, external sources, such as magnetic fields, ultrasound waves or laser beams, have been used for stimulating internal devices and triggering some therapeutic effects from outside, in a non-invasive way. The quest for targeted therapy has recently opened new opportunities for robotic technologies, which are used more and more as controllers for the delivery of drugs embedded in nanobiotech vectors and as solutions for making therapy really localized in the area of interest, enabling on-demand release kinetics and eliminating (or strongly limiting) side effects. This talk aims to present the above mentioned trends, with the support of specific examples coming from the speaker experience and her collaboration network. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Max Planck Intelligent Systems Colloquium, Stuttgart

Bacterial chemotaxis – principles and applications

Bacterial chemotaxis is one of the most thoroughly studied signaling systems in biology. We are interested in understanding the properties of Escherichia coli chemotaxis as an example of evolutionary optimized system, which has been selected for such properties as high sensitivity, signal integration, rapid response and robustness. We are further making use of the plasticity of the chemotaxis system to design sensitive and versatile biosensors that can detect a wide range of chemical compounds. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

3574 1447683032

Towards Flexible Intelligent Machines: System-level Challenges in Soft Robotics

With the ultimate objective of developing robots commonly used in real-world environments in close proximity to humans, we identify two key requirements; elasticity and accessibility. This talk will describe our research in the WPI Soft Robotics Laboratory on theoretical modeling, design, fabrication, actuation, sensing, and control solutions for soft robotic systems. Nature finds elegant solutions for many problems facing the robotics community, by harnessing mechanical compliance. Inspired greatly by biology, we envision future robotic systems to embrace mechanical compliance with bodies composed of soft and hard components as well as electronic and sensory infrastructure, such that robot maintenance starts to resemble surgery. A soft body offers safety and adaptability, which makes robots more suitable for use in a wide range of applications from human-robot interaction to search and rescue. Our approach to create flexible intelligent machines uses either soft materials or geometric arrangements of otherwise rigid elements. In the first track, we utilize fluidic actuation of elastomeric materials to generate organic deformations defined by geometry and embedded constraints in the substrate. Despite its advantages, mechanical compliance also violates many inherent assumptions in traditional robotics. Thus, a complete soft robot architecture requires new approaches to utilize accurate theoretical models that capture the nonlinear response of elastomeric materials, proprioception that provides rich sensory information while remaining flexible, and motion control under significant time delay. Our proposed solutions utilize nonlinear material models that predict motion and force output, integrated composite magnetic deformation and force sensing, and sliding mode feedback control of soft actuation to address each of these issues. In the second track, we create flexible robotic mechanisms in a cost- and time-effective manner, with the goal of building robots as easily as printing a document. We use common planar fabrication methods to create origami-inspired foldable bodies, and fabricate control circuits directly on this planar substrate. Utilizing a hierarchical development process of foldable robotic platforms as combinations of fundamental building-blocks, we can achieve arbitrary levels of complexity and functionality. Such designs make extensive use of foldable linkage mechanisms, which introduces a set of parameters that can be optimized for relevant task metrics and the crease pattern can be modified accordingly. This new philosophy of robot development offers improved design flexibility, ease of fabrication, cost-effectiveness, and lightweight robotic bodies compared to traditional systems. In this talk, I will discuss a variety of robots we developed as case studies, to demonstrate each of the new concepts we study towards the realization of complete soft robotic systems. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

3437 1458645017

Making Robots Learn

[mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Autonomous Robotic Systems and Multimodal Perception

To date, mobile robots have depended primarily on range sensors to explore and navigate in their environments. These sensors are used to build static 2D, 2.5D or 3D maps, or dynamic 4D maps that include information on spatial changes over time. However, as robots are deployed in increasingly more demanding missions in complex natural or artificial environments, they require richer world models with information beyond the spatial structure of their surroundings. This talk will provide an overview of research being conducted at the Autonomous Systems Program (CSIRO, Australia) and then discuss the use of multiple sensor modalities, including lidar, RGB, thermal, hyperspectral, and others, to build multi-property augmented world models (AWMs). Results will be shown from domains that include exploration and documentation of natural and cultural sites, as well as in situ sensing for agricultural, environmental and biodiversity applications. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Max Planck Intelligent Systems Colloquium, Tübingen

Max Planck Intelligent Systems Colloquium, Stuttgart

Surface Tension Effects in Microsystems

I will present my research on capillary forces and surface tension effects in microsystems (how to calculate capillary forces), including the Drobot (droplet robot) and ongoing research on thermocapillary micromanipulation (how to control surface tension effects). Pierre Lambert is Associate Professor at Université libre de Bruxelles. He obtained his MD in electromechanical engineering at ULB in 1998, his PhD in engineering sciences in the same university in 2004 and his habilitation in Université de Franche-Comté in 2010. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Towards Interactive Materials and Devices

The development of materials and devices with active and adaptive properties is one of the most profound challenges of today’s materials research. For exceeding the passive functionalities of existing materials, we more focus on programmable material properties, integration of active characteristics into interacting material systems, and related application in sustainable materials engineering and biotechnology. In this talk, I will present two example to show our latest progress in developing interactive materials and devices based on soft matter nanotechnology. The examples will include: (1) how natural biomaterials like proteins can be used for the fabrication of solid-state memory nanodevices, and (2) how to build soft electronic devices with advanced functions based on inorganic materials. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

2546 1444131036

Micro- and nano-scale technologies for applications in medicine and biology

Micro- and nano-scale technologies can have a significant impact on medicine and biology in the areas of cell manipulation, diagnostics and monitoring. At the convergence of these new technologies and biology, we research for enabling solutions to the real world problems at the clinic. Emerging nano-scale and microfluidic technologies integrated with biology offer innovative possibilities for creating intelligent, mobile medical lab-chip devices that could transform diagnostics and monitoring, tissue engineering and regenerative medicine. In this talk, we will present an overview of our laboratory's work in these areas focussed on applications in point-of-care and primary care settings including ovarian cancer detection from urine, rapid CD4 counts for global health, multiple pathogen detection with a focus on viral load from unprocessed whole blood. We will also introduce magnetic levitation methods for label free sorting of rare cells from whole blood. We will also review our work on 3-D biofabrication/bioprinting, and dynamic acoustic and magnetic systems for bottom-up tissue-construct assembly using cell encapsulating microscale hydrogels to engineer the 3-D cellular microenvironment. As an example, we will present a microfluidic platform, where flow induces a motile and aggressive phenotype in ovarian cancer nodules via increased epithelial-to-mesenchymal transition (EMT). These emerging technologies could shape our future creating broadly applicable platforms for scientific discovery, providing clinical solutions for resource-constrained settings in the developing world as well as for primary care settings in the developed world. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Nature-inspired innovation: from plants and animals to soft robots

Robots today are expected to operate in a variety of scenarios, being able to cope with uncertain situations and to react quickly to changes in the environment. In this scenario a strong relationship between nature and technology plays a major role, with the winning approach of evaluating natural systems to abstract principles for new designs. Bioinspired soft robotics is a worldwide known paradigm to develop new solutions for science and technology, giving way to a series of innovative robotic solutions assisting and supporting today’s society. Such biological principles traditionally originate from animal models for robots that can walk, swim, crawl, or fly. In this talk I will present some scientific and technological challenges and solutions coming from both animals and plants. In the animal paradigm a function is often related to an organ or compartment. Instead plants are networked, decentralized, modular, redundant, and resilient. Plants are able to move, control, sense, but they do in a different way with respect animals or other living beings. I will compare ideas, biological features, and technological translations coming from the two Kingdoms and related to areas of interest in robotics: movement, sensing and control. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Interplay between Merlin, Rac1 and Yap governs organ size and tumorigenesis

Neurofibromatosis type 2 (NF2) is caused by mutations of the tumor suppressor MERLIN/NF2. NF2 inactivation has been shown to lead to the activation of both YAP and RAC1. However, the relationship between RAC1 and the NF2-HIPPO-YAP pathway and how RAC1 contributes to NF2 tumorigenesis have not been elucidated. Using the mouse liver as a model system, we interrogated the genetic hierarchy of Nf2, Rac1 and Yap. Our findings showed that Yap and Rac1 function downstream of Merlin/Nf2 as two central signaling switches of a complex signaling network, which controls cell cycle progression, inflammation, DNA damage response, and senescence during NF2 tumorigenesis. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

2172 1435587656

Distributed Control of Modular Robots

A modular robot is a kind of robot built from mechatronic modules. Modules can be connected in many different ways and, hence, the same set of modules can be used to build a large variety of robots. Due to their modular structure, modular robots potentially are versatile, robust and cheap. However, to realise these potentials challenges in mechatronics design and control have to be addressed. In this talk I will give a brief introduction to modular robotics, the state-of-the-art in modular mechatronics, and an overview of the distributed control strategies explored. I will conclude with an outlook to the potential of employing soft robotics technologies in modular robotics. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Self-Powered Catalytic Nanomotors and Pumps

Self‐powered nano and microscale moving systems are currently the subject of intense interest. We have discovered that catalyst particles (both colloidal and molecular scale) generate sufficient mechanical force through substrate turnover to cause their own movement. Furthermore, the movement becomes directional through the imposition of a gradient in substrate concentration. The same catalyst particles, when anchored on a surface, pump ambient fluid directionally in the presence of the substrate. Possible applications involve sorting of catalysts based on activity, enhanced substrate channeling in tandem catalysis, enhanced mixing, and precisely controlled microfluidics. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Macroionic complex coacervation: concentration without perturbation of soft colloids

Complex coacervation is a liquid-liquid phase separation from solutions of oppositely charged polymers and colloids (e.g. polyelectrolytes, proteins, surfactant micelles). Brought to light more than 50 years ago, this field of science is experiencing a vigorous revival due to applications that include encapsulation, enzyme immobilization and separations, in biomedicine, food science and pharmaceutics. Modern methods of scattering, microscopy and rheology reveal to dynamic, equilibrium mesophases within optically clear coacervates, and help us understand why and how these structures self-assemble; while the earlier systems studied, comprising gelatin and other food proteins and polysaccharides, are superseded by more diverse macromolecules such as growth factors, hormones, disordered proteins, chitosan and hyaluronic acid. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

The dynamics of buckling of self-organizing contractile sheets

A central property of developing plant and animal tissues is their ability to form a diversity of folded patterns and adopt curved shapes through mechanical instabilities breaking the planar symmetry of growing epithelia. Folding of epithelia in animal tissues can be induced by imposing internal or external mechanical constraints that restrict the in-plane expansion of growing tissues. Developing tissues constitute examples of active matter and in many cases the underlying forces are contractile in nature. Here, we study the dynamics of initially homogenous, self-organizing thin elastic actin sheets that contract due to the activity of myosin motors. Motor-induced contraction starts at the system boundaries which later proceeds into the gel bulk resulting in spontaneous buckling of the sheet. The appearance of wrinkles is intimately linked to the development of nonlinear gradients in the gel density that develops spontaneously during contraction. The final patterns resemble wrinkled plant leaves. The buckling instability reported here is complementary to the buckling and folding of expanding tissues and offers a well-controlled system to study mechanically induced spontaneous shape transitions in active matter. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

The physics of touch shapes the early stages of neural somatosensory processing

The physics of contact are complicated and differ in fundamental ways from the physics of acoustics and optics. It therefore should be expected that the organization of the early stages of somatosensory processing be very different from that of equivalent stages in the other sensory modalities. The presentation will describe some salients facts regarding the physics of touch and will continue with the description of recent findings regarding the processing of time-evolving tactile inputs in second-order neurons in the cat. These results may call for a revision of the widely held idea that somatotopy is a strict organizational principle in the brain. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Soft Bio-inspired Robotics

Soft robotics is an emerging field of research that uses soft or compliant materials and elements to overcome the limitation of traditional robotics. Traditionally, robots have been used in an industrial environment with few unknown parameters. As more and more robots are used to interact with environments that are uncertain and vulnerable to change, a technology that can easily adapt to the changing environment is needed. Soft robotics deals with this issue by using soft and compliant elements in an intelligent way. Bio-inspired robotics is the field of robotics that leads the use of this technology. Nature has many examples where it achieves high performance with a soft intelligent design. Flytrap, for example, can close its leaves quickly by using bistability of the leaves and inchworm achieves adaptive gripping with its prolegs by using the buckling effect. These examples show that high performance can be achieved with a simple and minimum design. Also, wearable robots can be made more comfortable to wear, lightweight and small size by using soft robotics technology. Soft exoskeletons removes the frames from traditional exoskeleton robots and transmit forces to the body by directly connecting tendons, or soft actuators to the body. In this talk, I will give an overview of various soft bio-inspired robotic technologies and some of the robots that are being developed at SNU Biorobotics Laboratory. These soft robotic technologies will be helpful for robots that need to perform in rough or uncertain environments with a limited size, such as military robots and biomedical robots. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

From Perceptual Evidence to Large-Scale Visual Recognition Models

Recent progress in computer-based visual recognition heavily relies on machine learning methods trained using large scale annotated datasets. While such data has made advances in model design and evaluation possible, it does not necessarily provide insights or constraints into those intermediate levels of computation, or deep structure, perceived as ultimately necessary in order to design reliable computer vision systems. This is noticeable in the accuracy of state of the art systems trained with such annotations, which still lag behind human performance in similar tasks. Nor does the existing data makes it immediately possible to exploit insights from a working system - the human eye - to derive potentially better features, models or algorithms. In this talk I will present a mix of perceptual and computational insights resulted from the analysis of large-scale human eye movement and 3d body motion capture datasets, collected in the context of visual recognition tasks (Human3.6M available at http://vision.imar.ro/human3.6m/, and Actions in the Eye available at http://vision.imar.ro/eyetracking/). I will show that attention models (fixation detectors, scan-paths estimators, weakly supervised object detector response functions and search strategies) can be learned from human eye movement data, and can produce state of the art results when used in end-to-end automatic visual recognition systems. I will also describe recent work in large-scale human pose estimation, showing the feasibility of pixel-level body part labeling from RGB, and towards promising 2D and 3D human pose estimation results in monocular images.In this context, I will discuss perceptual, perhaps surprising recent quantitative experiments, revealing that humans may not be significantly better than computers at perceiving 3D articulated poses from monocular images. Such findings may challenge established definitions of computer vision `tasks' and their expected levels of performance. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Edible Electronics: Bioinspired materials and structures for next-generation ingestible devices

Ingestible electronic devices have the potential to obviate many of the challenges associated with chronic implants such as risk of infection, chronic inflammation, and costly surgical procedures. Current examples of ingestible electronics include edible cameras, biosensors, and integrated smart drug delivery systems. Ingestible devices have made great advances in the early detection and improved treatment of disease by using commodity polymers and off-the-shelf electronics. However, currently available materials fundamentally limit the scope and therapeutic potential of these devices. The potential clinical impact of ingestible electronics could be increased if the design toolbox features application-specific polymers and soft matter for this class of medical devices. This talk will describe recent advances in bioinspired materials for potential use in edible devices. Examples include flexible biodegradable elastomers as structural polymers and melanin-based pigments as materials for on-board energy storage. Structure-property-processing relationships in these biomaterials will be emphasized and prospective uses for these application-specific materials will be discussed. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Reflecting in and on the Gradient Domain

Image-based rendering has been introduced in the 1990s as an alternative approach to photorealistic rendering. Its key idea is to novel renderings by re-projecting pixels from nearby views. The basic approach works well for many scenes but breaks down if the scene contains “non-standard” elements such as reflective surfaces. In this talk, I will first show how we can extend image-based rendering to handle scenes with reflections. I will then discuss a novel gradient-based technique for image-based rendering that can intrinsically handle scenes with reflections. Bio: Michael Goesele studied computer science at Ulm University and the University of North Carolina at Chapel Hill. He then moved to the Max-Planck-Institut für Informatik (MPI) in Saarbrücken and received his doctorate degree in 2004. In 2005, he received a Feodor Lynen-Fellowship from the Alexander von Humboldt-Foundation to work as a postdoctoral researcher at the University of Washington, Seattle, USA. He joined TU Darmstadt in 2007 as assistant professor and became full professor in 2011. His research interests include capturing and modeling techniques for graphics and vision as well as computing on massively-parallel platforms. Dr. Goesele has published more than 70 papers in refereed international conferences and journal and received several awards including the Eurographics 2008 Young Researcher Award. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Cellular mechanotransduction in engineered extracellular matrices

The commitment of stem cells towards a specific lineage is influenced by many cues, which together constitute the cell microenvironment. One critical regulator is the extracellular matrix (ECM), which varies not only in composition, but also in physical properties such as stiffness. The impact of matrix stiffness on cell fate decisions has been studied intensively on 2D surfaces using synthetic hydrogels, but very little is known about stiffness sensing within 3D matrices. A main hurdle in 3D contexts is to isolate the role of ECM stiffness from other matrix properties, in particular degradability. If cells are fully encapsulated, changes in bulk stiffness also influence the amount of matrix crosslinks that a cell has to degrade in order to spread and interact with its surroundings, which impacts cell shape and function. Hence, there is a need to develop a fully synthetic hydrogel system that allows for independent control over the matrix properties governing cellular mechanosensing. We have developed a sugar-based hydrogel system that offers independent control over mechanical properties, adhesive ligand density and matrix degradation rates. The material can be processed under physiologic conditions rendering it suitable for cell encapsulation. The use of matrix metalloproteinase (MMP) cleavable peptide sequences as crosslinking units enables cellular matrix remodeling and variation of this sequence gives access to a range of degradation rates. Using this system, we study the impact of matrix stiffness and degradability on 3D cell spreading and angiogenic sprouting. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Industrial Robotics at Nanoscale

Current research activities in AMiR focus on the industrial robotics for nanohandling automation. The areas of research include nanohandling robots and systems; automated nanohandling methods; robot control methods for nanopositioning; fast vision feedback at nanoscale, etc. Prof. Fatikow introduces this new research field, the motivation, the key research problems and the applications at nanoscale. He specially addresses the current work on an automated nanohandling robot cell inside a scanning electron microscope (SEM). The latter serves as a powerful vision sensor and the work space for nanohandling robots equipped with application-specific tools. Major components – the piezo-driven nanohandling robots, the robot control system, the fast vision feedback – are discussed. Finally, current research projects in AMiR and related applications are outlined. They include e.g. automated assembly of nanophotonic structures, nanorobotic handling of graphene, automated assembly of nanorobotic tools, characterization of nanowires, etc. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Cell-Material Interfaces: Artificial Spores and Neurons on Nanotopographies

Nanometer-scaled objects (e.g., particles, rods, wires, and films) have the potential to elicit the biological responses from living cells and controllably perturb or modulate the cellular activities, as well as being utilized for probing, measuring, or analyzing the cellular information in vitro and in vivo. For example, in the neuron-material interfaces, the nanometric features accelerate the neurite outgrowth in vitro.1The accelerated neurite outgrowth has important biological implications, because the neurons encounter the hierarchical nano/microstructures of extracellular matrices in vivo, which are not presented faithfully in the homogeneous, flat environment of a 2D culture dish. Another example in the cell-material interfaces is the cytoprotection of individual living cells by forming nanometric coats on the cells.2 The cell-in-shell structures (“artificial spores”) show the enhanced tolerance against various external stresses in addition to the controlled cell division. The artificial shell degrades on-demand, mimicking the sporulation and germination processes found in nature. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Introduction to the Scenario Approach

The scenario approach is a broad methodology to deal with decision-making in an uncertain environment. By resorting to observations, or by sampling uncertainty from a given model, one obtains an optimization problem (the scenario problem), whose solution bears precise probabilistic guarantees in relation to new, unseen, situations. The scenario approach opens up new avenues to address data-based problems in learning, identification, finance, and other fields. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Forming and using magnets: Confronting the bacterial and the chemist’s views

Biomineralization is the formation of biological materials by living organisms. Even very primary organisms such as bacteria are able to produce inorganics with superior properties than man-made materials to fulfill a given function. Magnetotactic bacteria are a paradigm of such simple cells forming magnetic nanoparticles with magnetosomes, organelles dedicated to biomineralization. We will show how the magnetosomes are formed and how their ultrastructure, size, morphology, organization and orientation are biologically controlled in order to align with the Earth magnetic field lines for the search of particular oxygen conditions in aqueous environments. We will then compare these results with the reaction pathway we found for synthetic magnetite in the absence and presence of additives, and how the later can influence the properties of the nanoparticles that are formed as well as their assembly. Finally, we present how aggregates of such nanoparticles can be used as micro- and nanoswimmers and how those can be actuated by external magnetic fields. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Sensitivity, resolution and speed in detection and tracking of single biomolecules

Optical techniques are in high demand for the investigation of biomedical processes because they can be noninvasive, real-time and fast. In this talk, I present an overview of the recent advances in pushing the limits of sensitivity, resolution and speed in biological microscopy and how methods from laser spectroscopy, quantum optics and nanoscience have introduced a revolution in this area. In particular, I will show that photophysical improvements at low temperature can lead to optical resolution in the angstrom range, i.e. about one thousand times better than the diffraction limit. Next, I will discuss the need for fluorescence-free microscopy and how interferometric scattering detection (iSCAT) can be used for detecting individual biomolecules as small as 60 kDa in a direct and label-free fashion. The use of this method for very fast studies of diffusion and transport in lipid membranes is another important biophysical application that will be examined. If time allows, I will also discuss our recent work on trapping and manipulation of very small nanoparticles. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Images everywhere - computer vision with vehicle-mounted, airborne and tourist cameras

Abstract: I will present selected research projects of the Photogrammetry and Remote Sensing Group at ETH, including (i) 3D scene flow estimation for stereo video captured from a car; (ii) extraction of road networks from aerial images; and (iii) 3D reconstruction from large, unstructured (e.g. crowd-sourced) image collections. Bio: Konrad Schindler received the Diplomingenieur (M.Tech.) degree in photogrammetry from Vienna University of Technology, Austria, in 1999, and the Ph.D. degree from Graz University of Technology, Austria, in 2003. He has worked as a photogrammetric engineer in the private industry and held researcher positions in the Institute of Computer Graphics and Vision Department at Graz University of Technology, the Digital Perception Lab at Monash University, and the Computer Vision Lab at ETH Zurich. In 2009, he became Assistant Professor of Image Understanding at TU Darmstadt. Since 2010, he has been a tenured Professor of Photogrammetry and Remote Sensing at ETH Zurich. His research interests lie in the field of computer vision, photogrammetry, and remote sensing, with focus on image understanding and 3d reconstruction. He received several awards, including the U. V. Helava Award for the Best Paper in the ISPRS Journal of Photogrammetry and Remote Sensing 2008-2011 (with A. Ess, B. Leibe and L. Van Gool), and an honorable mention for the Marr Prize at ICCV 2013 (with C. Vogel and S. Roth). [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Examples of Machine Learning and Data Science at Facebook

Facebook serves close to a billion people every day, who are only able to consume a small subset of the information available to them. In this talk I will give some examples of how machine learning is used to personalize people’s Facebook experience. I will also present some data science experiments with fairly counter-intuitive results. Biography: Joaquin Quiñonero Candela is a Director of Engineering at Facebook, where he manages the Advertising Optimization organization. In the past he was a researcher at Microsoft Research in Cambridge UK, a postdoc at the Max Planck Institute in Tübingen and at the Fraunhofer Society in Berlin. Joaquin lives in Palo Alto, California, with his wife and three children. He is an avid paella cook and an obsessive long distance runner. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Nanophotonics meets Nanomechanics: Light force devices on a chip

Radiation pressure forces exerted by photons are generally considered to be too weak for real-world applications. The picture changes, however, when moving to nano-scale dimensions, at which considerable momentum transfer can result from illumination with light. Here I will describe a chip-scale framework in which optical forces can be exploited as an efficient driving and sensing mechanism for nano-scale resonators. Besides allowing for unprecedented measurement sensitivity, such opto-mechanical interactions provide a new paradigm for coupling nano-mechanical and nano-photonic components. I will provide an overview over progress on integrated opto-mechanics and also present recent results for chip-scale optical computing and mechanical data storage. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Approximate inference for stochastic differential equations

Stochastic differential equations (SDEs) arise naturally as descriptions of continuous time dynamical systems. My talk addresses the problem of inferring the dynamical state and parameters of such systems from observations taken at discrete times. I will discuss the application of approximate inference methods such as the variational method and expectation propagation and show how higher dimensional systems can be treated by a mean field approximation. In the second part of my talk I will discuss the nonparametric estimation of the drift (i.e. the deterministic part of the ‘force’ which governs the dynamics) as a function of the state using Gaussian process approaches. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

4D reconstruction in complex scenes, inverse rendering, advanced video editing

Even though many challenges remain unsolved, in recent years computer graphics algorithms to render photo-realistic imagery have seen tremendous progress. An important prerequisite for high-quality renderings is the availability of good models of the scenes to be rendered, namely models of shape, motion and appearance. Unfortunately, the technology to create such models has not kept pace with the technology to render the imagery. In fact, we observe a content creation bottleneck, as it often takes man months of tedious manual work by a animation artists to craft models of moving virtual scenes. To overcome this limitation, the research community has been developing techniques to capture models of dynamic scenes from real world examples, for instance methods that rely on footage recorded with cameras or other sensors. One example are performance capture methods that measure detailed dynamic surface models, for example of actors or an actor's face, from multi-view video and without markers in the scene. Even though such 4D capture methods made big strides ahead, they are still at an early stage of their development. Their application is limited to scenes of moderate complexity in controlled environments, reconstructed detail is limited, and captured content cannot be easily modified, to name only a few restrictions. In this talk, I will elaborate on some ideas on how to go beyond this limited scope of 4D reconstruction, and show some results from our recent work. For instance, I will show how we can capture more complex scenes with many objects or subjects in close interaction, as well as very challenging scenes of a smaller scale, such a hand motion. The talk will also show how we can capitalize on more sophisticated light transport models and inverse rendering to enable high-quality reconstruction in much more uncontrolled scenes, eventually also outdoors, and with very few cameras. I will also demonstrate how to represent captured scenes such that they can be conveniently modified. If time allows, the talk will cover some of our recent ideas on how to perform advanced edits of videos (e.g. removing or modifying dynamic objects in scenes) by exploiting reconstructed 4D models, as well as robustly found inter- and intra-frame correspondences. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Compressive Sensing and Beyond

The recent theory of compressive sensing predicts that (approximately) sparse vectors can be recovered from vastly incomplete linear measurements using efficient algorithms. This principle has a large number of potential applications in signal and image processing, machine learning and more. Optimal measurement matrices in this context known so far are based on randomness. Recovery algorithms include convex optimization approaches (l1-minimization) as well as greedy methods. Gaussian and Bernoulli random matrices are provably optimal in the sense that the smallest possible number of samples is required. Such matrices, however, are of limited practical interest because of the lack of any structure. In fact, applications demand for certain structure so that there is only limited freedom to inject randomness. We present recovery results for various structured random matrices including random partial Fourier matrices and partial random circulant matrices. We will also review recent extensions of compressive sensing for recovering matrices of low rank from incomplete information via efficient algorithms such as nuclear norm minimization. This principle has recently found applications for phaseless estimation, i.e., in situations where only the magnitude of measurements is available. Another extension considers the recovery of low rank tensors (multi-dimensional arrays) from incomplete linear information. Several obstacles arise when passing from matrices and tensors such as the lack of a singular value decomposition which shares all the nice properties of the matrix singular value decomposition. Although only partial theoretical results are available, we discuss algorithmic approaches for this problem. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Recent Advances and Opportunities in Aberration-Corrected Electron Microscopy

The development of aberration correction for electron microscopy has greatly improved our ability to characterize materials at the atomic scale. The TEAM project at the National Center for Electron Microscopy was launched to develop the next generation electron microscope with the goal of reaching a resolution of 0.5Å. The project was a collaborative effort between several DOE labs and commercial manufacturers. Since its successful conclusion in 2009, TEAM has contributed in numerous ways to advances in the field – by developing novel technologies, enabling new science and opening opportunities for innovation. This talk will briefly describe some of the new technologies such as electron-optical elements, stages, software and detectors that have now become broadly available. Their impact on materials characterization will be highlighted with examples of recent applications of the TEAM suite of instruments. Examples will include tomography, dynamics, observation of beam-sensitive materials, liquid cell microscopy, and high-precision measurement of atomic positions at grain boundaries in metallic and graphene bicrystals. Looking beyond the current state of the art, this talk will also outline some important new opportunities for further developments in electron microscopy. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Control of stick insect walking and beyond

Starting from a decentralized, reactive architecture able to control the behavior of a 20 DoF body, I will discuss how simple expansions of the controller could endow the system with higher level properties, i.e., with cognition (sensu McFarland and Bösser) and, time permitting, even with aspects of consciousness. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Low-rank dynamics

This talk reviews differential equations on manifolds of matrices or tensors of low rank. They serve to approximate, in a low-rank format, large time-dependent matrices and tensors that are either given explicitly via their increments or are unknown solutions of differential equations. Furthermore, low-rank differential equations are used in novel algorithms for eigenvalue optimisation, for instance in robust-stability problems. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Embedded Optimization for Nonlinear Model Predictive Control

This talk shows how embedded optimization - i.e. autonomous optimization algorithms receiving data, solving problems, and sending answers continuously - are able to address challenging control problems. When nonlinear differential equation models are used to predict and optimize future system behaviour, one speaks of Nonlinear Model Predictive Control (NMPC).The talk presents experimental applications of NMPC to time and energy optimal control of mechatronic systems and discusses some of the algorithmic tricks that make NMPC optimization rates up to 1 MHz possible. Finally, we present on particular challenging application, tethered flight for airborne wind energy systems. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Autonomous and non-autonomous dynamics of spin Hall auto-oscillations

A novel type of spin-torque nano-oscillators driven by pure spin current generated via the spin Hall effect - a spin Hall oscillator (SHO) - has been demonstrated. The SHO generates single-mode coherent auto-oscillations and combines local driving by pure spin current with enhanced spin-wave radiation losses. Counter-intuitively, the presence of radiation losses enables excitation of a single mode auto-oscillation and suppresses the nonlinear processes that prevent auto-oscillation by redistributing the energy between different modes. The study of the effects of external microwave signals on the SHO is also reported. It is shown that SHO can be efficiently synchronized by applying a microwave signal at approximately twice the frequency of the auto-oscillation, which opens additional possibilities for the development of novel spintronic devices. It is found that the synchronization exhibits an apparent threshold determined by magnetic fluctuations pumped above their thermal level by the spin current, and is significantly influenced by the nonlinear self-localized nature of the auto-oscillatory mode. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Towards Lifelong Learning for Visual Scene Understanding

The goal of lifelong visual learning is to develop techniques that continuously and autonomously learn from visual data, potentially for years or decades. During this time the system should build an ever-improving base of generic visual information, and use it as background knowledge and context for solving specific computer vision tasks. In my talk, I will highlight two recent results from our group on the road towards lifelong visual scene understanding: the derivation of theoretical guarantees for lifelong learning systems and the development of practical methods for object categorization based on semantic attributes. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Video Segmentation

Compared to static image segmentation, video segmentation is still in its infancy. Various research groups have different tasks in mind when they talk of video segmentation. For some it is motion segmentation, some think of an over-segmentation with thousands of regions per video, and others understand video segmentation as contour tracking. I will go through what I think are reasonable video segmentation subtasks and will touch the issue of benchmarking. I will also discuss the difference between image and video segmentation. Due to the availability of motion and the redundancy of successive frames, video segmentation should actually be easier than image segmentation. However, recent evidence indicates the opposite: at least at the level of superpixel segmentation, image segmentation methodology is more advanced than what can be found in the video segmentation literature. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Combining Sequential Analysis with Machine Learning for Solving Time-constrained Vision Problems

Computer vision problems often involve optimization of two quantities, one of which is time. Such problems can be formulated as time-constrained optimization or performance-constrained search for the fastest algorithm. We show that it is possible to obtain quasi-optimal time-constrained solutions to some vision problems by applying Wald's theory of sequential decision-making. Wald assumes independence of observation, which is rarely true in computer vision. We address the problem by combining Wald's sequential probability ratio test and AdaBoost. The solution, called the WaldBoost, can be viewed as a principled way to build a close-to-optimal “cascade of classifiers” of the Viola-Jones type. The approach will be demonstrated on four tasks: (i) face detection, (ii) establishing reliable correspondences between image, (iii) real-time detection of interest points and (iv) model search and outlier detection using RANSAC. In the face detection problem, the objective is learning the fastest detector satisfying constraints on false positive and false negative rates. The correspondence pruning addresses the problem of fast selection with a predefined false negative rated. In interest point problem we show how a fast implementation of known detectors can obtained by Waldboost. The “mimicked” detectors provide a training set of positive and negative examples of interest ponts and WaldBoost learns a detector, (significantly) faster than the providers of the training set, formed as a linear combination of efficiently computable feature. In RANSAC, we show how to exploit Wald's test in a randomised model verification procedure to obtain an algorithm significantly faster than deterministic verification yet with equivalent probabilistic guarantees of correctness. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

State-Space Representation of Gaussian Processes for Regression and Efficient Inference in Latent Force Models

Gaussian process regression is a non-parametric Bayesian machine learning paradigm, where instead of estimating parameters of fixed-form functions, we model the whole unknown functions as Gaussian processes. Gaussian processes are also commonly used for representing uncertainties in models of dynamic systems in many applications such as tracking, navigation, and automatic control systems. The latter models are often formulated as state-space models, where the use of non-linear Kalman filter type of methods is common. The aim of this talk is to discuss connections of Kalman filtering methods and Gaussian process regression. In particular, I discuss representations of Gaussian processes as state-space models, which enable the use of computationally efficient Kalman-filter-based (or more general Bayesian-filter-based) solutions to Gaussian process regression problems. This also allows for computationally efficient inference in latent force models (LFM), which are models combining first-principles mechanical models with non-parametric Gaussian process regression models. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Characterization of Gold Nanoparticles Functionalized with SAMs and Proteins

Nanomaterials used in biomedical diagnostic applications must exhibit well-controlled surface properties to achieve optimum performance in complex biological or physiological fluids. There is a need for development of rigorous and detailed methods for quantitative surface characterization of nanomaterials used in biomedical applications. Gold nanoparticles (AuNPs) ranging in diameter from 14 to 40 nm were functionalized with carboxylic acid, amine and ethylene glycol self-assembled monolayers (SAMs) were used in these studies. The surface chemistries of various chain length (C6 to C16) carboxylic acid terminated self assembled monolayers (COOH-SAMs) on AuNPs and flat Au surfaces were characterized with XPS, ToF-SIMS, TEM, FTIR and LEIS. As the AuNPs diameter decreased and SAMs chain length increased, the XPS atomic C/Au ratio on the surface increased and the ToF-SIMS intensity ratio of C1-4HxOy ions/Au-containing-ions increased. These trends showed that surface curvature had an effect on the XPS and ToF-SIMS measurements. Special XPS data analysis methods were developed to accurately analyze functionalized AuNPs. Simulated Electron Spectra for Surface Analysis (SESSA) was used to simulate the experimental XPS results as a function of take-off angle for COOH-SAMs on flat-Au surfaces. Quantities such as SAM density, thickness, surface roughness and instrumental parameters were tuned in SESSA to optimize agreement between SESSA and experimentally determined compositions. SESSA results showed that the thicknesses of the C16 COOH-SAM on 14nm AuNPs was 1.85 nm, which was 0.3 nm thinner that the same SAM on a flat Au surface. The presence of a thin 1.5Å CH2-contamination overlayer was detected on both samples. Low energy ion scattering (LEIS) experiments produced similar results (1.8 nm SAM thickness on the 14nm AuNPs). Results for oligo(ethylene glycol) (OEG) functionalized AuNPs showed that the type of end group (OH vs. OCH3) doesn’t have a significant effect on the SAM thickness and structure, but the size of the AuNP does. The C11 alkyl portion of the thiol molecules were well ordered on all surfaces (flat, 14nm and 40nm). In contrast, the four OEG units in the thiol molecules were better ordered and densely packed on the 40nm AuNPs compared to the 14nm AuNPs. LEIS measurements showed OEG SAMs had a thickness of 2.0 nm on the 14nm AuNPs compared to 2.6 nm on the 40nm AuNPs. Protein G was immobilized onto the HO-terminated OEG SAMs via carbonyl diimidazole chemistry. On flat Au surfaces XPS showed a monolayer of Protein G was covalently immobilized with little non-specific adsorption. On AuNPs a monolayer of Protein G could also be immobilized, but more non-specific adsorption was detected. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Online Spot Volatility-Estimation and Decomposition with Nonlinear Market Microstructure Noise Models

(joint work with Jan. C. Neddermeyer) A technique for online estimation of spot volatility for high-frequency data is developed. The algorithm works directly on the transaction data and updates the volatility estimate immediately after the occurrence of a new transaction. Furthermore, a nonlinear market microstructure noise model is proposed that reproduces several stylized facts of high frequency data. A computationally efficient particle filter is used that allows for the approximation of the unknown efficient prices and, in combination with a recursive EM algorithm, for the estimation of the volatility curve. We neither assume that the transaction times are equidistant nor do we use interpolated prices. We also make a distinction between volatility per time unit and volatility per transaction and provide estimators for both. More precisely we use a model with random time change where spot volatility is decomposed into spot volatility per transaction times the trading intensity - thus highlighting the influence of trading intensity on volatility. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Multi-View Perception of Dynamic Scenes

The INRIA MORPHEO research team is working on the perception of moving shapes using multiple camera systems. Such systems allows to recover dense information on shapes and their motions using visual cues. This opens avenues for research investigations on how to model, understand and animate real dynamic shapes using several videos. In this talk I will more particularly focus on recent activities in the team on two fundamental components of the multi-view perception of dynamic scenes that are: (i) the recovery of time-consistent shape models or shape tracking and (ii) the segmentation of objects in multiple views and over time. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Dissecting and reconstructing the microtubule cytoskeleton in vitro

The microtubule cytoskeleton is critical for a vast variety of essential processes in eukaryotic cells, including cell division and differentiation. How the microtubule cytoskeleton organises itself in space and how its dynamic properties are controlled are fascinating questions. The facts that the architecture and dynamic properties of this filament network can be observed by fluorescence microscopy make it an ideal object for the study of the design principles of a complex protein interaction network. Here we will present in vitro experiments aimed at dissecting microtubule cytoskeleton functioning by reconstructing it from purified components. Examples will be presented of how molecular mechanisms can be elucidated by reverse engineering of this major intracellular filament network. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Out-of-equilibrium materials constructed from molecular switches and superparamagnetic nanoparticles

Living organisms are the most prominent examples of systems self-assembled under far-from-equilibrium conditions. Inspired by nature, we design new materials that are capable of existing only as long as an external source of energy is supplied. In this talk, I will introduce molecular switches and superparamagnetic nanoparticles as the key building blocks of dynamically self-assembling materials. These new materials enable applications as diverse as light-controlled catalysis or manipulating nonmagnetic objects with the help of magnets. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Simulation in physical scene understanding

Our ability to understand a scene is central to how we interact with our environment and with each other. Classic research on visual scene perception has focused on how people "know what is where by looking", but this talk will explore people's ability to infer the "hows" and "whys" of their world, and in particular, how they form a physical understanding of a scene. From a glance we can know so much: not only what objects are where, but whether they are movable, fragile, slimy, or hot; whether they were made by hand, by machine, or by nature; whether they are broken and how they could be repaired; and so on. I posit that these common-sense physical intuitions are made possible by the brain's sophisticated capacity for constructing and manipulating a rich mental representation of a scene via a mechanism of approximate probabilistic simulation -- in short, a physics engine in the head. I will present a series of recent and ongoing studies that develop and test this computational model in a variety of prediction, inference, and planning tasks. Our model captures various aspects of people's experimental judgments, including the accuracy of their performance as well as several illusions and errors. These results help explain core aspects of human mental models that are instrumental to how we understand and act in our everyday world. They also open new directions for developing robotic and AI systems that can perceive, reason, and act the way people do. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Aperiodic Superlattices in Functional Oxide Nanowires

We are facing to develop alternative renewable energy sources, because of the limited fissile fuels and their impact on the environment. Functional oxide nanowires are expected to play an important role in scavenging waste heat and converting it into electricity. Some complex superlattices in nanostructures contain periodic compositional and structural features, typically on the nanometer scale, making them promising materials for thermoelectric applications. In this presentation, I will discuss our recently discovered a novel method to produce M2O3(ZnO)n polytypoid nanowires (M=In, Ga, Fe,..) by a facile solid state diffusion process. This is in agreement with the theoretical prediction that it is possible to increase the material-dependent figure of merit, zT, by using low dimensional materials, attributed to electronic band structure changes and enhanced interface phonon scattering. Atomic resolution HAADF imaging is used to perform a detailed structural analysis on the M2O3(ZnO)n nanowires, unambiguously determined the location of indium within the structure and to evaluate lattice strain and the presence of defects. Based on this analysis we propose that the superlattice structure is generated through a defect-assisted process. One of the greatest advantages of this novel synthesis is the ability to tune the nanoscale features of the polytypoid wires by simply adjusting the amount of metal precursor. We also performed a quantitative analysis of the change in superlattice inclusion density and periodicity with metal deposition. Compare to ZnO nanowires, these new oxide thermoelectric nanostructures exhibited almost three orders of magnitude increase in efficiency. This will enable future studies on structure-dependent thermoelectric properties and possibly lead to further enhancements in thermoelectric efficiency. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Biomimetics -the hopeful synthesis

In the last 20 years or so biology has emerged as a practical source of design concepts at all levels. Some of the ideas are reflections of pre-existing ideas from engineering, although the biology can open up novel applications and sophistications, and some are totallz novel in concept. There are also some amusing stories of misunderstanding (myth-understanding?) of principles and history. The lecture will start by involving you in a simple method to encourage creative and original thought and present a number of case histories taken from engineering, textiles, robotic medical devices, architecture and desing and a concept for a complete design and manufacturing process for biomimetic design. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Towards Visual Scene Understanding - Articulated Pose Estimation and Video Description

This talk will highlight recent progress on two fronts. First, we will talk about a novel image-conditioned person model that allows for effective articulated pose estimation in realistic scenarios. Second, we describe our work towards activity recognition and the ability to describe video content with natural language. Both efforts are part of a longer-term agenda towards visual scene understanding. While visual scene understanding has long been advocated as the "holy grail" of computer vision, we believe it is time to address this challenge again, based on the progress in recent years. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Identity Preserving Multi-People Tracking through Linear Programming

In this talk, I will show that, given probabilities of presence of people at various locations in individual time frames, finding the most likely set of trajectories amounts to solving a linear program that depends on very few parameters. This can be done without requiring appearance information and in real-time, by using the K-Shortest Paths algorithm (KSP). However, this can result in unwarranted identity switches in complex scenes. In such cases, sparse image information can be used within the Linear Programming framework to keep track of people's identities, even when their paths come close to each other or intersect. By sparse, we mean that the appearance needs only be discriminative in a very limited number of frames, which makes our approach widely applicable. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Max Planck Intelligent Systems Colloquium, Stuttgart

Design and Fabrication of Optical MEMS for Metrology

Micro Electro Mechanical Systems (MEMS) are well known especially through their use as accelerometers and gyrometers. These compact systems enable measurement of important parameters, such as acceleration, temperature, pressure, etc. Because of their size, a number of devices can easily be incorporated into a single system allowing multiple simultaneous measurements. As they are fabricated using technology similar to that used for VLSI circuits, the cost of the MEMS devices is also low. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

The Human Manifold - On the predictability of human online behaviour and its consequences

A growing proportion of human activities, such as social interactions, entertainment, shopping, gathering information, and learning, are now mediated by digital services and devices. Such digitally mediated behaviors can easily be recorded and analyzed, fueling the emergence of computational social science together with new services such as personalized search engines, recommender systems, and targeted online marketing. In this talk, I will discuss to what degree human behavior is predictable from online records, and I will demonstrate that, by using basic machine learning methods, it is possible to predict a wide range of personal attributes including sexual orientation, ethnicity, religious and political views, personality traits, intelligence, and happiness to a surprising degree of accuracy. I will discuss the ramifications of these findings for enhancing services in the context of personalization of Web search in Bing and address the resulting challenges for user privacy based on the principles of transparency and control. Time permitting I will talk about the ML Confidential project, which combines machine learning and homomorphic encryption in an attempt to reconcile prediction and privacy. The work on the predictability of human behaviour is in collaboration with Yoram Bachrach, Pushmeet Kohli (MSRC), and Milad Shokouhi (Bing), as well as Michal Kosinski and David Stillwell at the Psychometrics Centre of the University of Cambridge. The ML Confidential Project is in collaboration with Kristin Lauter and Michael Nährig in the Crypto group in MSR Redmond. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Damage and Repair in Bone: The Role of Microstructure and Cells

Bone sustains damage (microcracking) as a result of the forces applied during daily activities. Sometimes these small cracks growth to a large size, causing stress fractures. But normally, microdamage is effectively controlled by two factors: (i) the unique microstructure of bone, and; (ii) repair actions carried out by living cells. This talk describes work done in my research group and elsewhere to improve our understanding of these two factors. Bone is a quasi-brittle material which improves its essentially low toughness by using a complex hierarchical microstructure incorporating several toughening mechanisms acting at different length scales. We have shown that two concepts are useful in understanding these phenomena: the theory of critical distances (TCD) and the statistical analysis of large populations of microcracks. Without the constant actions of bone cells, bones would fail by fatigue cracking within less than one year. Our work has shed some light on the way in which osteocytes living inside the bone matrix are able to detect the presence of microcracks and signal to other cell types to organise repair activities. As part of this work, we were the first to demonstrate that cells can undergo fatigue failure during cyclic loading. Further understanding of microdamage and repair is important in order to combat bone diseases such as osteoporosis and to prevent and treat sports injuries in humans and animals. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Perceptual Grouping using Superpixels

Perceptual grouping played a prominent role in support of early object recognition systems, which typically took an input image and a database of shape models and identified which of the models was visible in the image. When the database was large, local features were not sufficiently distinctive to prune down the space of models to a manageable number that could be verified. However, when causally related shape features were grouped, using intermediate-level shape priors, e.g., cotermination, symmetry, and compactness, they formed effective shape indices and allowed databases to grow in size. In recent years, the recognition (categorization) community has focused on the object detection problem, in which the input image is searched for a specific target object. Since indexing is not required to select the target model, perceptual grouping is not required to construct a discriminative shape index; the existence of a much stronger object-level shape prior precludes the need for a weaker intermediate-level shape prior. As a result, perceptual grouping activity at our major conferences has diminished. However, there are clear signs that the recognition community is moving from appearance back to shape, and from detection back to unexpected object recognition. Shape-based perceptual grouping will play a critical role in facilitating this transition. But while causally related features must be grouped, they also need to be abstracted before they can be matched to categorical models. In this talk, I will describe our recent progress on the use of intermediate shape priors in segmenting, grouping, and abstracting shape features. Specifically, I will describe the use of symmetry and non-accidental attachment to detect and group symmetric parts, the use of closure to separate figure from background, and the use of a vocabulary of simple shape models to group and abstract image contours. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

The role of regularization in System Identification

The classical system identification setup with parameter estimation in selected model structures is first reviewed. Then the possibility to use large and unstructured model sets (high order FIR and ARX models) are studied. It is necessary to curb the flexibility by regularization. It turns out that this gives viable alternatives to the classical methods, and has links to current trends in machine learning. The role and difference between L2 and L1 penalties are discussed and illustrated. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Losses, Divergences and Deficiencies ... A Continuing Story

I will present some new results on loss functions and their connections to divergences, focussing on the multiclass case. I will characterise "proper composite" losses (which are the composition of a proper loss with a link function) and explain why they are important for machine learning. I will show how multiclass proper losses are equivalent to a generalisation of f-divergences that take multiple arguments (that is that jointly measure the "distance" between k different distributions simultaneously. I will also spell out the relationship between this work and the classical comparison of experiments, and in particular present a new formula for the deficiency distance between two multiclass experiments. Deficiency, introduced by Le Cam some 50 years ago, is a means by which the classical notion of a sufficiency (as in "sufficient statistic") can be extended to a notion of approximate sufficiency. Finally, if time allows, I will outline how both losses and divergences can more elegantly be defined in terms of a single convex set, rather than in terms of functions. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Depth, You, and the World

Consumer level depth cameras such as Kinect have changed the landscape of 3D computer vision. In this talk we will discuss two approaches that both learn to directly infer correspondences between observed depth image pixels and 3D model points. These correspondences can then be used to drive an optimization of a generative model to explain the data. The first approach, the "Vitruvian Manifold", aims to fit an articulated 3D human model to a depth camera image, and extends our original Body Part Recognition algorithm used in Kinect. It applies a per-pixel regression forest to infer direct correspondences between image pixels and points on a human mesh model. This allows an efficient “one-shot” continuous optimization of the model parameters to recover the human pose. The second approach, "Scene Coordinate Regression", addresses the problem of camera pose relocalization. It uses a similar regression forest, but now aims to predict correspondences between observed image pixels and 3D world coordinates in an arbitrary 3D scene. These correspondences are again used to drive an efficient optimization of the camera pose to a highly accurate result from a single input frame. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Efficient Algorithms for Semantic Scene Parsing

Developing autonomous systems that are able to assist humans in everyday's tasks is one of the grand challenges in modern computer science. Notable examples are personal robotics for the elderly and people with disabilities, as well as autonomous driving systems which can help decrease fatalities caused by traffic accidents. In order to perform tasks such as navigation, recognition and manipulation of objects, these systems should be able to efficiently extract 3D knowledge of their environment. In this talk, I'll show how Markov random fields provide a great mathematical formalism to extract this knowledge. In particular, I'll focus on a few examples, i.e., 3D reconstruction, 3D layout estimation, 2D holistic parsing and object detection, and show representations and inference strategies that allow us to achieve state-of-the-art performance as well as several orders of magnitude speed-ups. Bio: Raquel Urtasun is an Asssistant Professor at TTI-Chicago a philanthropically endowed academic institute located in the campus of the University of Chicago. She was a visiting professor at ETH Zurich during the spring semester of 2010. Previously, she was a postdoctoral research scientist at UC Berkeley and ICSI and a postdoctoral associate at the Computer Science and Artificial Intelligence Laboratory (CSAIL) at MIT. Raquel Urtasun completed her PhD at the Computer Vision Laboratory, at EPFL, Switzerland in 2006 working with Pascal Fua and David Fleet at the University of Toronto. She has been area chair of multiple learning and vision conferences (i.e., NIPS, UAI, ICML, ICCV), and served in the committee of numerous international computer vision and machine learning conferences. Her major interests are statistical machine learning and computer vision, with a particular interest in non-parametric Bayesian statistics, latent variable models, structured prediction and their application to semantic scene understanding. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Neue Effekte bei Varistoren aus Zinkoxid: Lokalisierung des Stromtransportes, Asymmetrie des Verhaltens der Korngrenzen und Druckabhängigkeit der elektrischen Eigenschaften

Varistoren sind polykristalline Halbleiterbauelemente mit einer stark nicht-linearen Strom- Spannungs- Kennlinie, die als Überspannungsschutz von elektrischen und elektronischen Systemen Verwendung finden. Sie werden aus geeignet dotiertem (Bi, Al, Co, Mn, ...) ZnO-Pulvern durch Sintern hergestellt. Dabei werden Ionen in der Korngrenzen eingebaut und längs der Korngrenzen bilden sich Raumladungszonen (Doppel- Schottky Barrieren; DSBs), welche die Ladungen in den Korngrenzen kompensieren und eine elektrostatische Barriere für den Elektronentransport bilden. Unterhalb einer charakteristischen Spannung, der Schaltspannung VB, sind die Varistoren gute elektrische Widerstände. Bei höheren Spannungen (V > VB ) werden heiße Elektronen und durch Stoßionisation auch Elektronen-Loch Paare erzeugt. Die Löcher wandern zur Korngrenze, wo sie mit den dort gebunden Elektronen rekombinieren und so die Potentialbarre abbauen. Dadurch steigt der Strom I überproportional mit der Spannung V an (I ~ V α) an. Der Exponent kann Werte über 100 (α>100) erreichen. Experimentelle Untersuchungen an Varistoren zeigen eine Reihe überraschender Effekte. So kann es z.B. bei Erreichen der Schaltspannung zu einer Lokalisierung des Stromtransportes kommen. Das "stromtragende" Volumen kann dabei auf Bruchteile von Prozenten des Bauteilvolumens schrumpfen. Dies führt dann zu einer starken lokalen Erwärmung. Es kann auch zum lokalen Aufschmelzen und zur Zerstörung des Bauteiles kommen. Sehr kleine Bauteile, so wie sie in der modernen Mikroelektronik als Überspannungsschutz verwendet werden, zeigen häufig eine starke Streuung der elektrischen Eigenschaften und je nach Vorzeichen des angelegten Feldes unterschiedliche elektrische Eigenschaften. Dies konnte durch komplementäre experimentelle Techniken (lock-in Thermographie, keramographische Präparation relevanter Korngrenzen, scanning surface potential microscopy, EBSD, …) auf das Verhalten einzelner Korn- Kornübergänge zurückgeführt werden. Durch Messen der Strom-Spannungs-Kennlinie der relevanten Korngrenzen (4-Punkt Technik) konnte gezeigt werden, daß deren Verhalten vom Vorzeichen des elektrischen Feldes abhängt und daß die Orientierungsbeziehungen zwischen den Körnern dabei eine Rolle spielen. Weiters konnte gezeigt werden, daß der elektrische Widerstand von Bauteilen sich unter mechanischem Druck um nahezu das 1000-fache verändern kann. Die Höhe der Potentialbarrieren hängt von der Zahl der Ladungsträger in den Korngrenzen ab. Wegen der Piezoelektrizität der ZnO-Kristalle werden diese von den Eigenspannungen in der Keramik und den Orientierungsbeziehungen zwischen den Körnern beeinflußt. Auch die Druckabhängigkeit des Widerstandes ist auf die Piezoelektrizität zurückzuführen. Diese Zusammenhänge können durch ein 3-dimensionales Netzwerkmodell mit einem realistischen Gefüge beschrieben werden, in dem die Mikroeigenspannungen aufgrund der anisotropen thermischen Ausdehnung der Körner nach dem Sintern und die dadurch bewirkten Ladungsverschiebungen in den Korngrenzen berechnet werden. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Is Ignorance Bliss with Imprecise Probabilities?

When uncertainty is represented with sets of probabilities -- so called "Imprecise Probability" [IP] theory -- rather than with a single probability distribution, what becomes of the basic result from Subjective Expected Utility [SEU] theory that cost-free information has non-negative value? That is, under SEU theory, where uncertainty is represented by a single probability distribution and the decision rule is to maximize expected utility -- and in the absence of moral hazard-- the rational agent, YOU, should postpone a decision in order first to acquire new, cost-free information. SEU theory advises YOU to value cost-free information as having non-negative instrumental value for making better decisions. If the new information might alter your decision, then ex-ante, such information has positive value to YOU. Only if YOU will decide the same, regardless of the new information, it has no ex-ante added value for YOU. I review necessary and sufficient conditions, within IP theory, for new evidence to dilate a set of probabilities. Evidence from an experiment dilates a set of probabilities when, for each possible experimental outcome, the uncertainty in the updated IP set of probabilities increases. The IP increase in uncertainty happens because the lower and upper conditional probabilities, given the new evidence, are more extreme than their corresponding lower and upper unconditional probabilites. That is, with dilation the new evidence is certain to increase IP-uncertainty in the sense that the updated IP intervals of probability become strictly larger compared with the IP intervals of probability prior to updating. For a variety of IP-decision rules, an experiment that causes dilation produces cost-free information with negative IP value. For instance, with dilation, the IP decision rule to maximize minimum expected utility leads YOU to give the new evidence negative value: YOU strictly prefer to make the decision without updating first, contrary to the basic SEU result summarized above. The same results about dilation can be interpreted as showing when new, shared information among several Bayesians is sure to lead them to diverging posterior probabilities, in contrast with the familiar (asymptotic) results about the merging of posterior probabilities with shared evidence. With dilation, the priors swamp the data in the posteriors, and not the other way around! [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Signaling at the physical limit – chemotaxis in sperm

Eggs attract sperm by releasing chemical factors. Sperm of the sea urchin Arbacia punctulata respond to the binding of a single molecule of chemoattractant with a brief Ca 2+ spike. We have studied the molecular basis for this exquisite sensitivity. The chemoattractant receptor, a guanylyl cyclase (GC) which synthesizes cGMP, is densely packed covering almost the entire surf ace of the flagellum (ca. 1,000,000 copies/flagellum). The density is similar to that of rhodopsin in the disc membrane. Binding of chemoattractant to the trimeric receptor is controlled by negative cooperativity, which allows sperm to operate over a 100,0 00 - fold range of concentrations without becoming saturated. Chemoattractant and receptor form a relatively long - lived complex that synthesizes only a few molecules of cyclic GMP/s, i.e. receptor signaling is of low gain. However, cyclic GMP activates in a non - cooperative fashion a unique pseudo - tetrameric cyclic nucleotide - gated K + channel that displays nanomolar cGMP sensitivity. A single chemoattractant molecule causes the cell to hyperpolarize by about 1 - 2 mV. The small hyperpolarization activates a suff icient number of Ca v channels. At rest, the receptor GC is multiply phosphorylated. Signaling is terminated by inactivation of the receptor within 0.3 s via rapid dephosphorylation of the GC. The changes in [Ca 2+ ] control the flagellar beat and thereby the swimming path. At rest, sperm swim on regular circles. In a chemical gradient, the periodic stimulation while swimming in circles is translated into a periodic modulation of the path curvature. Alternating periods of high and low curvature give rise to a motility pattern of drifting circles that move towards the source of the chemoattractant. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Object Recognition by Hierarchical Learning Machines

I will describe work in my group over the last 2 ye ars developing a theory of visual cortex and of deep learning architectures of the co nvolutional type. I will describe the theoretical consequences of a simple assumption: th e main computational goal of the feedforward path in the ventral stream – from V1, V 2, V4 and to IT – is to discount image transformations, after learning them during d evelopment. The initial assumption is that a basic neural operation consists of dot pr oducts between input vectors and synaptic weights – which can be modified by learnin g. I will outline theorems showing that a multi-layer hierarchical architecture of dot -product modules can learn in an unsupervised way geometric transformations of image s and then achieve the dual goals of invariance to global affine transformation s and of robustness to deformations. These architectures develop to be automatically inv ariant to transformations of a new object, achieving the goal of recognition with very few labeled examples. The theory should apply to a range of hierarchical architectur es such as HMAX, convolutional “deep learning” networks and related feedforward mo dels of sensory systems and extend their power. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Interactive Variational Shape Modeling

Irregular triangle meshes are a powerful digital shape representation: they are flexible and can represent virtually any complex shape; they are efficiently rendered by graphics hardware; they are the standard output of 3D acquisition and routinely used as input to simulation software. Yet irregular meshes are difficult to model and edit because they lack a higher-level control mechanism. In this talk, I will survey a series of research results on surface modeling with meshes and show how high-quality shapes can be manipulated in a fast and intuitive manner. I will outline the current challenges in intelligent and more user-friendly modeling metaphors and will attempt to suggest possible directions for future work in this area. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

3D vision in a changing world

3D reconstruction from images has been a tremendous success-story of computer vision, with city-scale reconstruction now a reality. However, these successes apply almost exclusively in a static world, where the only motion is that of the camera. Even with the advent of realtime depth cameras, full 3D modelling of dynamic scenes lags behind the rigid-scene case, and for many objects of interest (e.g. animals moving in natural environments), depth sensing remains challenging. In this talk, I will discuss a range of recent work in the modelling of nonrigid real-world 3D shape from 2D images, for example building generic animal models from internet photo collections. While the state of the art depends heavily on dense point tracks from textured surfaces, it is rare to find suitably textured surfaces: most animals are limited in texture (think of dogs, cats, cows, horses, …). I will show how this assumption can be relaxed by incorporating the strong constraints given by the object’s silhouette. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Where are the intersitital carbon and nitrogen atoms in steels?

The distribution of interstitial carbon and nitrogen atoms in steels has been a subject of investigation for many years. The scientific significance of this is considerable, because these interstitial species control a number of phenomena of immense engineering importance, including strain ageing, embrittlement and tempering response. The experimental technique of atom probe microanalysis permitted direct observation of carbon and nitrogen atom distributions on the atomic scale for the first time. An overview of this work will be presented. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

What Make Big Visual Data Hard?

There are an estimated 3.5 trillion photographs in the world, of which 10% have been taken in the past 12 months. Facebook alone reports 6 billion photo uploads per month. Every minute, 72 hours of video are uploaded to YouTube. Cisco estimates that in the next few years, visual data (photos and video) will account for over 85% of total internet traffic. Yet, we currently lack effective computational methods for making sense of all this mass of visual data. Unlike easily indexed content, such as text, visual content is not routinely searched or mined; it's not even hyperlinked. Visual data is Internet's "digital dark matter" [Perona,2010] -- it's just sitting there! In this talk, I will first discuss some of the unique challenges that make Big Visual Data difficult compared to other types of content. In particular, I will argue that the central problem is the lack a good measure of similarity for visual data. I will then present some of our recent work that aims to address this challenge in the context of visual matching, image retrieval and visual data mining. As an application of the latter, we used Google Street View data for an entire city in an attempt to answer that age-old question which has been vexing poets (and poets-turned-geeks): "What makes Paris look like Paris?" [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Deep Gaussian Processes

In this talk we will introduce deep Gaussian process (GP) models. Deep GPs are a deep belief network based on Gaussian process mappings. The data is modeled as the output of a multivariate GP. The inputs to that Gaussian process are then governed by another GP. A single layer model is equivalent to a standard GP or the GP latent variable model (GPLVM). We perform inference in the model by approximate variational marginalization. This results in a strict lower bound on the marginal likelihood of the model which we use for model selection (number of layers and nodes per layer). Deep belief networks are typically applied to relatively large data sets using stochastic gradient descent for optimization. Our fully Bayesian treatment allows for the application of deep models even when data is scarce. Model selection by our variational bound shows that a five layer hierarchy is justified even when modelling a digit data set containing only 150 examples. In the seminar we will briefly review dimensionality reduction via Gaussian processes, before showing how this framework can be extended to build deep models. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Solving Problems with Visual Analytics: The Role of Visualization and Analytics in Exploring Multidimensional Data

Never before in history data is generated and collected at such high volumes as it is today. As the volumes of multidimensional data available to businesses, scientists, and the public increase, their effective use becomes more challenging. Visual analytics seeks to provide people with effective ways to understand and analyze large multidimensional data sets, while also enabling them to act upon their findings immediately. It integrates the analytic capabilities of the computer and the abilities of the human analyst, allowing novel discoveries and empowering individuals to take control of the analytical process. In the visual analysis process, it is not obvious what can be done by automated analysis and what has to be done by interactive visual methods. In dealing with massive data, the use of automated methods is mandatory, but there is also a wide range of problems where the use of interactive visual methods is necessary. The talk presents the potential of visual analytics and discusses the role of automated versus interactive visual techniques. A variety of application examples ranging from document analysis over network security to molecular biology illustrate the exiting potential of visual analysis techniques but also their limitations. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Learning to Optimize with Confidence

In many applications, we have to make decisions with unknown rewards. Often, the number of choices is extremely large (possibly infinite). We can model this problem as maximising an unknown reward function that is expensive evaluate from a small number of noisy samples. A key challenge is trading off exploration (estimating the function) and exploitation (maximizing the estimated function). In this talk, I will present recent progress towards this fundamental problem. I will show how a simple confidence-guided sampling rule attains near-minimal regret for a large class of reward functions, modeled as samples from Gaussian processes or having low RKHS norm. I will further demonstrate how our approach allows to scale up through parallelization, incorporate context to more quickly solve related tasks, and address multi-objective tradeoffs. I will illustrate the approach in several real-world applications. Applied to experimental design for protein structure optimization, our approach enabled engineering of active P450 enzymes that are more thermostable than any previously made by chimeragenesis, rational design, or directed evolution. [mehr]

Max Planck Intelligent Systems Colloquium, Stuttgart

Atom Probe Tomography – Basics, Data Analysis and Application to the Analysis of Phase Transformations

Atom probe tomography (APT) is now well established as a routine method in the materials characterization toolbox. Since its availability is becoming increasingly wide-spread, it seems useful to give an introductory overview over the method. The first part of this talk will comprise a description of the method with its capabilities and limitations, focusing on practical aspects. After stating the kind of problems that can be tackled with APT, specimen preparation methods will be presented as well as all steps necessary to prepare, perform and post-process a typical APT experiment. Moreover, it will be shown how specialized data analysis techniques, including cluster analysis and field evaporation simulations, can be used to extract valuable information beyond colorful pictures from APT results. The second part of the talk will be dedicated to APT experiments performed to investigate phase transformations in metals. Examples will include the analysis of precipitation kinetics in electrical steel, of diffusion-controlled austenitisation in maraging steel as well as segregation phenomena in pearlitic steel. The talk will conclude with an outlook on future improvements in APT experimentation and data analysis. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Unsupervised Learning for Computer Vision - Generative Models of Visual Occlusions, Efficient Parameter Optimization, and Neural Encoding

The ability to process and interpret data is a core functionality studied in the fields of Machine Learning, Computer Vision and Computational Neuroscience. The key for data interpretation can hereby be found in the data’s underlying regularities which are ultimately caused by the highly regular nature of the processes that have generated the data. Recent years have seen an increased interest in approaches that build sophisticated representations of data to accomplish a variety of tasks ranging from pattern detection, object recognition to cue integration or action planning. For many tasks it is crucial that the learned representations appropriately reflect the true data structure. In my talk I will discuss different approaches for the representation of data with a focus on unsupervised learning and visual data. I will start with standard models such as sparse coding or independent component analysis but will immediately criticise the assumptions made by these approaches. For visual data, the assumption of linear component superposition is unrealistic, for instance, because occlusions are not modeled. This motivates the investigation of alternative approaches which do take visual occlusions into account. In contrast to prominent lines of research, e.g., on restricted Boltzmann machines or autoencoders, I focus on fully interpretable directed graphical models. The crucial problem to solve for such models is parameter optimization which is analytically and computationally intractable. Building-up on novel variational approximations in combination with massively parallel computing, I show how such challenges can be overcome. The application of the resulting algorithms to different types of data then allows for a demonstration of the models' capabilities in different settings. Finally, I discuss the implications of the new models and their results for neural encoding. In particular, I ask what impact occlusions in visual data may have on the neural encoding in primary visual cortex. I show results of standard linear and occlusive models for visual data, and discuss their differences and similarities to in vivo recordings as well as to other models in the field. [mehr]

Max Planck Intelligent Systems Colloquium, Tübingen

Safe Learning: learning the learning rate in Bayesian, MDL and PAC-Bayesian inference

Standard Bayesian and MDL inference can behave badly if the model under consideration is wrong: in some simple settings, the posterior fails to concentrate even in the limit of infinite sample size. We introduce a test that can tell from the data whether we are heading for such a situation. If we are, we can adjust the learning rate (equivalently: make the prior lighter-tailed, or penalize the likelihood more) in a data-dependent way. The resulting "safe" estimator continues to achieve good rates with wrong models. It can also be interpreted as a method for automatically determining the learning rate in PAC-Bayesian settings. For example, when applied with classification models, the safe estimator achieves the optimal rate for the Tsybakov exponent of the underlying distribution. The safe estimator is based on empirical mixability, which generalizes a central concept in worst-case online prediction due to Vovk. Thus, safe estimation connects the three main paradigms in learning: Bayesian inference, (frequentist) statistical learning theory and (worst-case) on-line prediction. [mehr]

 
loading content