Metal Nanostructures for Optical Sensing and Signaling
Jim Adleman, Demetri Psaltis

Abstract. The aim of this research is to develop devices based upon two dimensional arrays of metallic nanoparticles, with an optical signatures that are tunable and can measure changes their environment. We have synthesized silver nanoparticles of 3-6 nm in diameter. We have measured resonant scattering from solutions and 2D arrays of these particles throughout the visible spectrum. The resonance of these particles is due to the motion of the ‘free’ electrons in the cluster.

We attempt to modify the shape of this resonance by distorting the shape of the electron cloud of the particle with an external field. To study this effect we spin coat silver nanoparticles on to clear conductive substrates in order to apply large fields both along the direction of propagation and the direction of polarization of light that passes through our devices. Non-linear interaction between nanoparticles which can be tuned by applied fields would make it possible to switch electromagnetic energy confined to a nanometer scale at optical frequencies. This would be very useful in the design of optical switches for computing, and arrays of nanoparticle based sensors that could be used to measure chemical or physical changes in a given environment.

We also are attempting electrical tuning of the metal insulator transition in silver nanoparticles. When a lattice of sufficiently identical nanospheres is compressed so that the electron spillout from individual crystals overlap, the electron states become delocalized across the whole lattice. This gives the lattice the characteristics of a thin metal film. We propose to use external fields to re-localize these electrons to single sites in the lattice. This would allow the film to switch between a metallic state with a flat absorption curve and an insulating state with a resonant absorption curve. (full report)

Modeling Swarm-Based, Distributed Robotic Manipulation
William Agassounon, Kjerstin Easton, and Alcherio Martinoli
Collaborators: Joel Burdick, Kristina Lerman, Wulfram Gerstner

Abstract. We developed a macroscopic modeling methodology for swarm-based, distributed robotic manipulation. The methodology is well-suited for nonspatial metrics as it does not take into account robots’ trajectories or the spatial distribution of objects in the environment. The strength of the proposed models is that they have been built up incrementally, with matching between models and embodied simulations (and sometimes, real robot experiments) verified at each step as new complexity was added. Precise heuristic criteria based on geometrical considerations and systematic tests with one or two embodied agents prevent the introduction of free parameters into the model. Two concrete case-studies were considered. The first case-study, referred to as the aggregation experiment, is a non-collaborative manipulation concerned with gathering and clustering small objects initially scattered in an enclosed arena. The other case-study is involves strictly collaborative manipulation and is referred to as the stick-pulling experiment, as the robots’ task is to collaborate to pull sticks out of holes in the arena floor. Results show that the proposed approach delivers quantitatively accurate predictions, in particular for nonspatial metrics related to both the aggregation and stick-pulling processes, and constitutes a computationally efficient tool. The simplicity of the modeling methodology suggests that it is easily applicable to other experiments characterized by different agent capabilities and individual control algorithms. (full report)

Networks, Evolution, Science & Neural Systems
Alex Bäcker

Abstract. Recent times have seen the advent of large amounts of data on networks of diverse kinds, from the WWW and citation networks to protein and gene expression networks. Part of my work has been aimed at extracting insight out of these massive collections of data. We show, for example, that recent years have seen an expansion in the memory of science and a homogenization of citation distributions. In parallel, I have been developing mathematical methods to extract information from multi-neuron recordings of brain activity. More generally, I am addressing a variety of open questions at the interface of biology, math and computation. (full report)

CMOS Imager with Embedded Analog Early Image Processor
Christophe Basset, Bedabrata Pain (JPL), Pietro Perona

Abstract. We are developing a computational CMOS imager with integrated early image processing general-purpose filter. The goal of this collaborative work with the Jet Propulsion Laboratory is to produce a single chip serving as a camera able to pre-process the image in real-time through a filter chosen by the user, allowing an efficient implementation of a variety of computationally intensive applications such as autonomous navigation, object avoidance or intercept, real-time target tracking and recognition. (full report)

Spike Based Saliency Detection
Ulrik Beierholm, Pietro Perona

Abstract. Trying to quickly ascertain which parts of a visual scene is most relevant for a recognition task and then focusing on each of these areas, is an economical use of processing power known to be employed in the human visual system. Most models for saliency detection however are too slow to explain the performance of the biological system. We are currently working on implementing a fast neuronal spike based saliency detector model based on rank order coding. (full report)

Fly Flight Simulator to Study Visual and Rotational Stimuli
John Bender, Michael Dickinson, Pietro Perona

The fly flight arena was designed (not by me!) to explore the connections between the different sensory modalities that fruit flies use to control their flight. The fly is glued to a metal post mounted in the center of a cylindrical arena. The walls of this cylinder are made out of 11,340 LEDs which are controlled in real time by a computer. (Flies have poor spatial resolution, estimated at 5°, but very fast temporal resolution - around 200 Hz. Human vision has spatial resolution of about 1/30th degree and temporal resolution around 20 Hz.) (full report)

Encoding of Depth in Parietal Reach Region (PRR)
Rajan Bhattacharyya, Richard Andersen

Technological developments in the past decade have accelerated the pace of research in brain computer interfaces. Multiple research groups across the country are pursuing this area of research as a possible solution to spinal cord injury. The Andersen lab at Caltech specializes in studying brain areas in the parietal cortex, which is associated with vision and motor planning, and in particular the Parietal Reach Region (PRR) which encodes the plan for the next intended reach movement, which is markedly different than the approach taken by other research groups which are using the motor cortex as the source of control signal. The Cortical Prosthetic Project at the Andersen lab has multiple research areas, including the development of an implantable chip to read signals from the parietal cortex, development of computational models for the neural signals involved, development of an online decoding algorithm for the intended movements, and finally the implementation of the real time control of a robotic arm through a brain computer interface.

This project seeks to investigate the encoding of depth by PRR neurons by carrying out experiments that in essence characterize the system. The first experiment will involve training non-human primates to maintain fixed eye positions while reaching to targets at various locations in three dimensional space. The second experiment will have the primates vary eye positions, however maintain fixed reach locations. Subsequently, we will investigate the neural mechanism by which PRR neurons encode the intended three dimensional reach location and develop a computational model to simulate the process. Lastly, we will augment the online decoding algorithm that is under development to decode PRR signals from implanted arrays in non human primates to control a robotic arm in real time to make reaches to locations in three dimensional space. (full report)

Reward Expectancy in Dorsomedial Frontal Cortex
of the Macaque Monkey
M. Campos, B. Breznen, and R. A. Andersen

Abstract. We recorded neural activity from the dorsomedial frontal cortex of two macaque monkeys during the performance of memory guided and object based saccade tasks. Target locations in both tasks were identical, and event defined intervals could be readily compared across tasks. In about 75% of the recorded neurons we observed a burst of activity during the interval following the instructed saccade in both tasks. The majority (65%) of these neurons also showed a shift in the onset time of this burst from one task to the other. The burst occurred immediately after the target-acquiring saccade in the object based task, but with a ~250ms delay in the memory guided task. The timing of the burst corresponded to the appearance of the visual feedback that indicated to the monkey that he successfully completed the task. Furthermore, in successful trials the burst terminated with the delivery of the reward, but in error trials, in which the monkey attempted the proper saccade but was not rewarded, the burst was sustained for up to 2 seconds. We interpret these results to mean that the burst activity in these cells reflects an expectation of a reward, and that it persists until the reward is obtained.

Robotics Facilitation in Spinal Learning
Lance Cai, Andy Fong, Joel Burdick and V. Reggie Edgerton

Each year, 11,000 Americans suffer spinal cord injury. Victims of severe spinal cord injury may suffer symptoms as severe as paraplegia, quadriplegia, and death. Currently we have no means of restoring locomotor function to patients who have suffered severe neural tissue damage resulting from spinal cord injury. While ideal treatments for such injuries involve regenerating the damaged tissues or developing compensatory neural connections, these options are not yet feasible. For patients who have lost the ability to walk, however, promising studies indicate that properly conducted, systematic motor training may help them walk again. (full report)

Reward Expectancy in Dorsomedial Frontal Cortex
of the Macaque Monkey
M. Campos, B. Breznen, and R. A. Andersen

We recorded neural activity from the dorsomedial frontal cortex of two macaque monkeys during the performance of memory guided and object based saccade tasks. Target locations in both tasks were identical, and event defined intervals could be readily compared across tasks. In about 75% of the recorded neurons we observed a burst of activity during the interval following the instructed saccade in both tasks. The majority (65%) of these neurons also showed a shift in the onset time of this burst from one task to the other. The burst occurred immediately after the target-acquiring saccade in the object based task, but with a ~250ms delay in the memory guided task. The timing of the burst corresponded to the appearance of the visual feedback that indicated to the monkey that he successfully completed the task. Furthermore, in successful trials the burst terminated with the delivery of the reward, but in error trials, in which the monkey attempted the proper saccade but was not rewarded, the burst was sustained for up to 2 seconds. We interpret these results to mean that the burst activity in these cells reflects an expectation of a reward, and that it persists until the reward is obtained.

Holographic Time-Resolved Imaging of Plasma Generated by High-Intensity Laser Pulses
Martin Centurion, Demetri Psaltis

Abstract. We study the formation and time-evolution of plasma generated in air by high intensity femtosecond pulses. We recorded holographic images of the plasma filaments on a CCD camera, which allowed us to reconstruct the phase change induced by the plasma on a probe. The distribution of the free electrons in the plasma is derived from the phase change, revealing multiple filaments and their breakup and recombination. We also demonstrated the capability of this holographic technique for capturing the time evolution of the plasma generation process by capturing a sequence of images of the filaments in a single-shot experiment. (full re port)

Path-Planning for Feature-Recognition and Classification using Information Theoretic Methods
Tim Chung, Joel Burdick, Richard Murray

Abstract.This project investigates the role of information-theoretic techniques in cooperative multi-agent systems. These techniques are used to govern the path planning of agents to optimally classify features of interest by improving the quality of the measurements. Sensor measurements are assumed to be in the presence of noise. We consider issues associated with distributed systems such as sensor fusion of information and formation control of relative vehicle locations. The objective is to articulate the theory underlying the relationship between sensing tasks and cooperative control. (full report)

Distributed Exploration and Coverage
Nikolaus Correll, Kjerstin Easton, Alcherio Martinoli, and Joel Burdick
Collaborators: Jonathan Witt, Edmond Wong (NASA Glenn Center)

Abstract. The aim of this project is to formulate an efficient exploration and coverage algorithm for a swarm of mobile agents. We present a completely distributed algorithm relying on agents endowed with identical controllers. The controller for the individual agent is realized through a hybrid approach using deliberative planning together with reactive behavior for collision avoidance. To exchange information about task progress the agents exploit a cellular decomposition of the environment. Coverage is performed using a grid-based algorithm (the Spanning Tree Coverage algorithm). Interaction between the agents is constrained to decentralized line-of-sight communication with limited range. The algorithm has been proved regarding completeness and its performance has been systematically investigated using an embodied simulator. (full report)

Decomposition of Human Motion into Dynamics Based
Primitives with Application to Drawing Tasks
Domatilla Del Vecchio, Richard Murray, Pietro Perona

Abstract. Using tools from dynamical systems and systems identification we develop a framework for the study of primitives for human motion, which we refer to as movemes. The objective is understanding human motion by decomposing it into a sequence of elementary building blocks that belong to a known alphabet of dynamical systems. We develop a segmentation and classification algorithm in order to reduce a complex activity into the sequence of movemes that have generated it. We test our ideas on data sampled from five human subjects who were drawing figures using a computer mouse. Our experiments show that we are able to distinguish between movemes and recognize them even when they take place in activities containing an unspecified number of movemes. (full report)

The Stochastic Nature of Single Neurons
Kamran Diba, Christof Koch

Our labs have been very active in furthering our understanding of the biophysical noise in neocortical pyramidal cells. The Hebrew University group traveled to California in March, and Dr. Kamran Diba traveled twice to Jerusalem in April and August to discuss and advance our collaborative research. Theoretically, we have strengthened our understanding of the role of ion channels and synaptic vesicular release in determining the voltage noise fluctuations. Experimentally, we made more measurements under varied pharmacological conditions. We also developed a method for quantifying instrumental noise, and we began measuring the input impedance of the cell with zap currents. We presented a poster at the Society for Neuroscience meeting in November. We are presently working to understand some of the low-frequency noise features that we recently uncovered. (full report)

Human Motion Detection and Classification
Claudio Fanti, Pietro Perona

Abstract. We foresee a future in which machines autonomously interact with Humans in the surrounding environment. So far, very good results have been achieved in detecting the presence of Humans and labeling their body parts by means of graphical-models based algorithms. We unavoidably have to deal with uncertainty and reasoning in absence of complete information. To that extent, we explore and enhance the state of the art in probabilistic inference and sampling techniques having the machines understanding human actions as a primary application. (full report)

Visuo-olfactory sensory fusion for flight behavior in flies
Mark Frye, Michael Dickinson

Over the past year I have used the support of this grant to study the neurobiological basis of multisensory flight control in flies. I have specifically focused on vision and olfaction and how feedback from these sensory modalities is integrated to coordinate complex spatiotemporal dynamics of search behaviors. Using a state-of-the-art stereo video system, I tracked freely flying flies within different sensory landscapes and found that visual expansion cues generated as flies approach vertical edges is required for odor localization (Fig. 5A). Using a 'virtual reality' tethered flight simulator, I examined the fine scale motor responses to visual expansion, odor, and both presented simultaneously. Our results show that during flight sensorimotor responses to odor are linearly superimposed upon visual responses (Fig. 5B). This is a remarkable finding because it suggests that – from an engineering perspective - the underlying neural processing for tracking multiple sensory cues is relatively simple. A parallel sensory-to-motor control architecture may be an evolutionary adaptation that imparts both the extraordinary flexibility and robustness exhibited by flies in diverse sensory landscapes. These results have culminated in one publication, presentations at two international meetings, and two more manuscripts to be submitted for publication this month. (full report)

Line Source Approximation Predicts Extra-Cellular Voltage for CA1 Neurons Recorded In Vivo
Carl Gold, Christof Koch, Darrell Henze, Gyorgy Buzsaki

Abstract. The Line Source Approximation (LSA) is a mathematical method for calculating the extracellular field from a 3-D distribution of membrane current sources. We investigate the use of the LSA combined with detailed compartmental modeling, including a model of the electrodes used, to predict the extracellular voltage waveform shape and magnitude resulting from the spiking activity of individual neurons. This provides an estimate of the maximal distance at which a neuron could be detected by an extracellular electrode. In order to tune the model we compare simultaneous intracellular and extracellular recordings of CA1 neurons recorded in vivo with model predictions for the same cells reconstructed and simulated. The approximate electrode position is estimated from the histologically determined track. We overcome the uncertainty regarding the values of biophysical parameters, such as the extra-cellular conductivity and the membrane Na+ conductance, by comparing the model and experimental results for numerous samples of the same class of neuron. Based upon comparisons with experimental data, we conclude that the compartmental model can accurately simulate the in vivo intracellular action potential and the LSA model can accurately simulate the extracellular fields of individual spiking neurons.

Fast Bayesian Support Vector Machine Parameter Tuning with the Nystrom Method
Carl Gold, Alex Holub

Abstract. We experiment with speeding up a Bayesian method for tuning the hyperparameters of a Support Vector Machine (SVM) classifier. The Bayesian approach gives the gradients of the evidence as averages over the posterior, which can be approximated using Hybrid Monte Carlo simulation (HMC). By using the Nystrom approximation to the SVM kernel, our method significantly reduces the dimensionality of the space to be simulated in the HMC. We show that this speeds up the running time of the HMC simulation from O(n^2) (with a large prefactor) to effectively O(n), where n is the number of training samples. We conclude that the Nystrom approximation has an almost insignificant effect on the performance of the algorithm when compared to the full Bayesian method, and gives excellent performance in comparison with other approaches to hyperparameter tuning.

The Involvement of the Anterior Cingulate Cortex in Novelty
Han C.J., Anderson D.J., & Koch, C.

The activation of the anterior cingulate cortex was previously shown to correlate with novelty detection. However, whether the anterior cingulate cortex is necessary to novelty detection is unclear. We set up a novelty object paradigm in mice. Mice were brought to the testing room in their home cage. A group of mice received a novel object (a corning 15 ml tube), a group received the same procedure including lifting the cage lid but not the object, and a group received nothing. We showed that the novel object readily induces the exploratory behaviors of the mouse directed towards the novel object, and cage lid lifting induces general exploratory behaviors. The sum of time that the group receiving the novel object and the group receiving the lid lifting spend in exploratory behaviors are equal, but the exploratory behaviors in the group that received the novel object are mostly directly to the object. c-fos mRNA was used as a surrogate marker to detect neuronal activation by in situ hybridization on brains from each group. Animals from each of the three groups were sacrificed 30 minutes after the first exposure of the stimulus. We discovered that there are more c-fos positive cells in the anterior cingulate cortex of the brain that received the novel object, compared with the other two groups. To answer the question whether the anterior cingulate cortex is necessary for novelty detection, a group of mice received excitotoxic lesions of the anterior cingulate cortex and another group received sham surgery. Behavioral experiments and analyses are being conducted to determine whether the lesions to the anterior cingulate cortex cause any exploratory behavioral changes directed to the novel object.

Applications of Carbonized Parylene for Sensor Technology
Ted Harder, Yu-Chong Tai

Abstract. Currently I am working on a new material, carbonized parylene. This new material provides a cheap easy and flexible way to micromachine carbon on a silicon substrate. This form of carbon has great potential for a large number of sensors. Currently we are investigating it as a humidity sensor, NO (nitrous oxygen) sensor and its application to a bolometer (infrared light/heat sensor).

During the last six months we have investigated the pyrolysis process by which we carbonize parylene and several of the fundamental material properties. We have found that the material is highly porous which means the film has a high amount of surface area. We have also characterized the temperature coefficient of resistance which can be used in both the bolometer and in high temperature heaters.

Several fabrication related challenges have been characterized and the process has been modified to improve the overall potential of the fabrication process.

Due to the importance of carbon various types of chemical sensing and its inertness, the ability to micromachine a form of carbon has implications for a wide variety of novel sensors.

Dynamic Recurrent Neural Networks for Pattern Recognition
Alex Holub, Gilles Laurent, Pietro Perona

We are investigating the computational properties of recurrent neural networks of binary artificial neurons. Our investigations are guided by recent work performed in the laboratory of Gilles Laurent which involves elucidating the underlying processing mechanisms in early olfactory processing. These physiological investigations indicate that the initial olfactory processing layer (in the locust the antennal lobe) consists of a dynamic recurrent neural network of excitatory and inhibitory units. The presentation of stimuli to the network results in stereotyped spatio-temporal neural firing patterns, with each unique stimulus presentation invoking a unique temporally-varying pattern of activity within the population of neurons. We have approximated the biological networks using recurrent networks with discrete binary neural elements. These non-linear networks exhibit chaotic behavior such that similar input patterns obtain very dissimilar network representations through the network dynamics. Similar pattern spreading characteristics have been observed in the initial processing networks of fish by members of the Laurent laboratory and it has been hypothesized that pattern spreading may be one computational benefit which the initial processing layer provides. (full report)

Athermal Holographic Filters
Hung-Te Hsieh, Demetri Psaltis, Yu-Chong Tai

Abstract. Holographic filters are used as optical sensors and in wavelength division multiplexing (WDM) filtering applications. Temperature dependence is a critical concern for telecommunications. We realize the design of an athermal holographic filter employing a thermally actuated MEMS mirror to compensate for the drift of Bragg wavelength due to changes of temperature. The center wavelength of our holographic filter is shown to remain constant from 21°C to 60°C. (full report)

Computational Modeling of Feature Inheritance
Whee Ky (Wei Ji) Ma

The proposition that reentrant interactions into the early visual system are necessary for visual awareness has lately been under close scrutiny. We examine this proposition in the context of a neuronal model which explains the phenomenology of feature inheritance. (full report)

Optimization and Generalization in Boosting
Ling Li, Yaser Abu-Mostafa

Abstract. The superior out-of-sample performance of AdaBoost has been attributed to the fact that it minimizes a cost function based on margin. In order to examine how the cost function, in and of itself, affects the out-of-sample performance, we apply several more sophisticated optimization techniques directly to the cost function. When the AdaBoost exponential cost function is optimized, our methods generally yield much lower cost and training error but higher test error, which implies that the exponential cost is vulnerable to overfitting. With the optimization power gained, we can adopt more "regularized" cost functions that have better out-of-sample performance but are difficult to optimize. Our experiments demonstrate that with suitable cost functions, our methods can have better out-of-sample performance. (full report)

Rapid Natural Scene Categorization without Attention
Fei Fei Li, Rufin VanRullen, Christof Koch, Pietro Perona

Abstract. What can we see when we do not pay attention? While attention is not necessary for some detection tasks on simple synthetic stimuli, without it we are “blind” even to major aspects of a natural complex scene. It would thus appear that only visual tasks that have an explanation in the early stages of the visual system may be carried out without attention. We report on a complex visual task that requires no attention. Our subjects can rapidly detect animals in briefly presented natural scenes while simultaneously performing another visual task that demands full attention. By comparison, they are unable to discriminate large ‘T’s from ‘L’s in the same conditions. We conclude that attention may not be necessary for some visual tasks that are associated with ‘high level’ cortical areas. (full report)

Object Categorization: Unsupervised One-Shot Learning
Fei-Fei Li, Rob Fergus, Pietro Perona

Abstract. Learning visual models of object categories notoriously requires thousands of training examples; this is due to the diversity and richness of object appearance which requires models containing hundreds of parameters. We present a method for learning object categories from just a few images (1 - 5). It is based on incorporating "generic'' knowledge which may be obtained from previously learnt models of unrelated categories. We operate in a variational Bayesian framework: object categories are represented by probabilistic models, and "prior'' knowledge is represented as a probability density function on the parameters of these models. The "posterior'' model for an object category is obtained by updating the prior in the light of one or more observations. Our ideas are demonstrated on four diverse categories (human faces, airplanes, motorcycles, spotted cats). Initially three categories are learnt from hundreds of training examples, and a "prior'' is estimated from these. Then the model of the fourth category is learnt from 1 to 5 training examples, and is used for detecting new exemplars a set of test images. (full report)

Volume Holographic Filters for Spectroscopic Identification of Substances
Zhenyu Li, Demetri Psaltis

We use volume holography to create spectrally specific, selective filters for the identification of substances such as toxic or explosive materials. The identification method is spectroscopy (such as IR or Raman spectroscopy) where the identity of molecules is found in the detailed absorption or emission spectra. Volume holographic filters are able to improve the sensitivity and speed of the measurement by detecting multiple absorption (or emission) spectral lines of the given substance simultaneously. The operation is based on the Bragg selectivity and multiplexing ability of volume holograms. It’s well known that within the dynamic range of the holographic recording medium, multiple holograms can be superimposed, or multiplexed, in the same volume, which makes it possible to construct a holographic filter whose wavelength selectivity curve (spectral response curve) is matched precisely to the absorption spectrum of a given substance. In order to achieve this, a special recording exposure schedule must be carefully designed such that the strength and spectral bandwidth of individual hologram are matched precisely to those of the corresponding peak in the spectrum. With multiple peaks detected simultaneously, it’s expected the detection sensitivity and speed will be increased greatly compared with traditional methods, and the required data volume will decrease by several orders of magnitude, which makes it very attractive for remote sensing applications.. (full report)

Uncooled All-Parylene Bolometer
Matthieu Liger, Yu-Chong Tai

We present here a novel, low-cost uncooled parylene bolometer. The device is made of two layers of pyrolyzed parylene and a metal layer for interconnections. We demonstrate that high responsivity can be achieved by tailoring the electrical conductivity and the temperature coefficient of resistance (TCR) using different pyrolysis conditions for each parylene layer.
(full report)

Computational Modeling of Feature Inheritance
Whee Ky (Wei Ji) Ma

The proposition that reentrant interactions into the early visual system are necessary for visual awareness has lately been under close scrutiny. We examine this proposition in the context of a neuronal model which explains the phenomenology of feature inheritance. (full report)

Neuromechanical Design and Active Sensory Systems in Animals
Malcolm Maciver, Joel Burdick

The field of neuroethology has made tremendous progress in understanding the sensory processing that subserve natural behaviors. Much work remains, however, in obtaining an equally detailed and quantitative understanding of how the mechanics of animals subserve natural behaviors, and in particular, how sensory abilities complement an animal’s mechanical control and locomotory needs and characteristics. In addition to its basic science import, these issues have relevance to engineers seeking to emulate some of key advantages of animal neuromechanical design, such as high maneuverability, and high levels of sensory integration for executing behaviors under changing and uncertain conditions. In this work we study how motion and sensing are integrated in the weakly electric fish. (full report)

Parylene Technology for Mechanically Robust Neuro-Cages
Ellis Meng, Yu-Chong Tai, Jon Erickson, and Jerome Pine

Abstract. We present a novel process to produce parylene cages for the in vitro study of cultured neural networks. For the first time, a neuro-cage fabrication technology is demonstrated that is scalable to high density cage arrays and able to withstand the chemical and mechanical rigors of supporting cellular cultures for long-term study.
(full report)

Mismatch Reduction in an On-Chip Image Processing Chip
Performing Feature Detection
Ania Mitros, Christof Koch

Feature extraction is a first step for many existing computer vision algorithms. This computation is also often one of the most time- and resource-intensive steps because the same local computation must be performed at each pixel. To head towards a real-time, small-size, energy-efficient implementation, Pesavento implemented the Tomasi- Kanade feature extraction algorithm in silicon. Although each feature detector worked splendidly, transistor mismatch killed the performance of the array. I have been re-implementing the blocks of the feature detector with floating gate transistors within each to permanently program away the mismatch. I have implemented mismatch reduction in the photoreceptor and the multiplier; both are tested and function as desired.

Suppressive Effect of Sustained Low-Contrast Adaptation followed by Transient High-Contrast on Peripheral Target Detection
Farshad Moradi, Shinsuke Shimojo, Christof Koch

Filling-in can be induced by high-contrast edge adaptation, or after prolonged adaptation to a peripheral low-contrast object (Troxler fading). Adaptation to sustained low-contrast vs. adaptation to transient high-contrast suggests synergy between contrast and edge adaptation, but the possible interactions are not well understood. We observed that briefly increasing the contrast of a peripheral low-contrast object after a few seconds of strict fixation elicits disappearance of the object, resulting in perceptual filling-in of the location with the surround (Figure 1a). After a short time usually around one second the object reappears. Hence, following sustained adaptation to a low-contrast target, transient high-contrast stimulation can induce perceptual disappearance. (full report)

Object Recognition by Probabilistic Hypothesis Construction
Pierre Moreels, Michael Maire, Pietro Perona

Abstract. We present a probabilistic framework for recognizing objects in images of cluttered scenes. Hundreds of objects may be considered and searched in parallel. Each object is learned from a single training image and is modeled by the visual appearance of a set of features, as well as their position with respect to a common reference frame. The recognition process computes both the identity and position of objects in the scene by computing the best interpretation (or hypothesis) of the scene in the light of a database of known objects. A hypothesis pairs features in an input image either with features in the database or marks them as clutters. Each hypothesis may be scored in a principled way using a generative model of the image which is defined using the learned objects as well as a model for clutter. While the space of all possible hypotheses is enormously large, one may find the best hypothesis efficiently—we explore a couple of heuristics to do so. In our initial experiments our algorithm compares favorably with state-of-the-art recognition systems. (full report)

Control Algorithm for Movable Neuro-Probes
Zoran Nenadic

Abstract. The process of extracellular recording from animals cortex is rarely automated. Moreover, such a procedure requires a constant human supervision and could be very time consuming. Here we propose a new algorithm that automatically controls the position of a recording electrode, while maintaining a certain level of signal quality. (full report)

Holographic Spatial-Mode-Division-Multiplexing for
Fiber Optic Sensors
Eric Ostby, Demetri Psaltis

Abstract. Fiber optic sensors are currently used to measure temperature, pressure, strain, power, chemical concentrations and more [1]. Evanescent fiber optic sensing is the most popular. The evanescent tails of guided modes interact with the surrounding medium. Information about chemicals or perturbations there are obtained by measuring the change in mode power, polarization or delay. Key benefits of fiber optic sensors include its compact size, durability in extreme environments, low power requirements, and low cost.

Currently, fiber optic sensors do not have control over specific modes, only large groups [2]. For instance, it is desirable to launch significant power into higher order modes to increase the sensitivity of the instrument. But, only one-dimensional knowledge is possible with such limited schemes. Each spatial mode has a different fraction of its power traveling outside the fiber core. The penetration depth of each mode is different, and therefore provides two-dimensional accuracy in measurement. By comparing the power loss of several modes, radial information about concentration variations from the core can be calculated.

The goal of this project is to use a novel multiplexing technique to gain exact control over every spatial mode in optical fibers. Mode-division-multiplexing (MDM) uses the spatial modes present in optical fiber as an orthogonal basis. The spatial profiles of multiple modes are stored in a volume hologram. Individual modes are launched and detected with angle-multiplexed holograms. Therefore, accurate information of mode attenuation due to the surrounding medium is known. In addition to sensing applications, addressing the spatial modes of a multimode fiber (MMF) increases the bandwidth of an optical communication system [3]. Multiple modes in the transmission channel provide extra degrees of freedom, and hence greater capacity [4]. Presently, fiber optic communication systems do not use the spatial modes to carry information. Modal dispersion decreases the useable bandwidth of MMF links that do not address the multimode nature of the channel [5]. This project will also implement the MDM scheme to increase the bandwidth, and therefore, the speed in MMF communication systems. (full report)

Computational Modeling of Visual Attention Systems
Robert J. Peters, Asha Iyer, Christof Koch
Nathan Mundhenk, Laurent Itti

Abstract. We have continued to extend our biological model of bottom-up visual attention with several recently characterized retinal and cortical interactions that are known to govern human performance in certain visual tasks. We are testing the behavioral importance of these interactions by comparing our model's predictions against human eye movement data recorded with our infrared eyetracker. In the last year we have worked with three new model components: (1) short-range orientation interactions (for clutter reduction), (2) long-range orientation interactions (for contour facilitation), and (3) retinal filtering (for fovea vs. periphery effects). (full report)

Propelling Underwater Vehicles Using Vortex Ring Generation
Ann Marie Polsenberg, Joel Burdick

Abstract. As robots designed to operate underwater become more common, it is useful to look at ways to make them more efficient. Autonomous Underwater Vehicles (AUVs) carry their power source with them, so improving the efficiency of the vehicle will also increase the maximum duration time for missions that the vehicle can perform. One area in which efficiency is very important is the propulsion system. We propose that vortex ring generators may be a viable way to propel these vehicles. This idea stems from looking at aquatic animals, such as squid, which use this mechanism. Our work involves the modeling, design, construction and analysis of synthetic jets. The next step will be to design a small vehicle that uses these thrusters and to begin an investigation into the control of such a vehicle. (full report)

Monotonic Bernoulli Trials
Amrit Pratap, Yaser Abu-Mostafa, Pietro Perona

Abstract. When estimating a number of bernoulli variables which have a certain monotonicity constraint, if the number of samples for each variable is small, then the estimates will not satisfy the monotonicity constraint. Better performance is achieved by endorcing the monotonicity constraint on the estimation procedure. (full report)

Inter-stimulus Distance Effects in Visual Search
Lavanya Reddy, Rufin VanRullen, Christof Koch

Abstract. In a previous study, we showed that the attentional requirements of a task, as revealed by the dual-task paradigm, do not necessarily determine whether visual search will be parallel or serial. For example, natural scene categorization can be performed "preattentively" in a dual-task situation (i.e., a single scene containing animals can be discriminated from non-animal scenes even while attention is occupied elsewhere), and yet visual search for an animal scene among a number of non-animal scenes is a serial process. We interpreted these findings as follows: a task can be performed preattentively if there exist specific neuronal populations selective to the target and distractor categories, independent of the level of processing involved (from V1 to IT); when such selectivities exist, visual search is parallel only if the receptive fields of the relevant neurons do not significantly overlap. When receptive fields are too large, target and distractors compete within the same field and search is serial. It follows that search performance should improve if target and distractors can be separated enough to prevent them from falling into the same receptive field. We tested this prediction and found that for preattentive tasks that usually result in serial visual search (e.g., color-orientation conjunction discrimination, upright vs. inverted face discrimination), search performance improved as inter-stimulus distance was increased. For preattentive parallel tasks (color discrimination, orientation discrimination), the effect of increasing inter-stimulus distance was negligible. These results support the idea that for preattentive tasks, competition within the relevant receptive fields can affect visual search performance.

Modular Electronics for Rapid Development of Behavioral Stimuli
Michael Reiser, Michael Dickinson

Whereas flies use many sensory modalities, most of the behaviors we casually observe are dominated by visual control. For this reason, presenting controlled visual environment to tethered flies continues to be a powerful experimental paradigm. Most experiments have been done in simple arenas, either patterns attached to a rotating drum, or in recent years, using cylinders covered with LEDs. Conventional display technologies (LCDs, CRTs, etc.) can not be used as stimuli for insect experiments, because their refresh rates are typically several times slower than the flicker fusion rate of insect visual systems. LEDs are used because they can be rapidly refreshed, which is necessary to maintain the illusion of motion. We have designed modular panels of 64 LEDs each, which can be snapped together to ‘tile’ an experimental environment with controllable displays. The panels are individually addressed and communicated with via a rapid serial interface. The panels have been designed to be extremely bright (with the added flexibility of individual pixel programmable brightness control), allowing experimentation over a broad range of behaviorally relevant stimuli conditions. The panels are controlled via a microprocessor controller which, for most experiments, will not require a computer in the loop, significantly reducing the infrastructure necessary for experiments. This technology allows an experimenter to build a visual arena with a customized geometry in a matter of hours.
(full report)

Vision as a Compensatory Mechanism for Disturbance Rejection in Upwind Flight
Michael Reiser, Michael Dickinson, Sean Humbert, Richard Murray

For several decades the visuo-motor control system of flies has been extensively studied. However, recent results have cast new light on many long standing assumptions about the operation of the flight control system. In this project we seek to demonstrate that through a faithful model of the fly's behavior, it is possible to provide some context within which controlled behavioral assays can be interpreted. (full report)

Decoding Neuroprosthetic Control Signals from Human Parietal Cortex
Daniel Rizzuto, Richard Andersen

Recent work in macaques has shown that different areas of posterior parietal cortex are specialized for planning hand and eye movements (1; 2), and that it is possible to use recordings from these areas to predict the direction of the planned movement (3). Preliminary studies from our group have taken the first step toward identifying the human homologue of the macaque parietal reach region (PRR), which is responsible for planning hand movements (4). However, it is still unknown if neural activity in human PRR exhibits the same spectral characteristics as that in the macaque. To address this question we are working with human participants who have chronically implanted electrodes placed on the surface of cortex and within deep brain regions, often in partial cortex. Recording taken from these participants while they execute delayed reaches allow us to acquire high signal-to-noise intracranial EEG (iEEG) activity from cortical areas during motor planning. Analysis of this neural activity is aimed at determining which properties of the signal can be used to decode and predict planned movement.

Additionally, in order for human PRR to serve as a substrate for neuroprosthetic control signals it must be resistant to pathological reorganization after cortico-spinal tract (CST) injury, an issue which is still a matter of debate. To address this, we have begun using fMRI to examine differences in motor planning activity in quadriplegic patients compared to normal participants. This comparison will allow us to see to what degree the activity in these areas degenerates after CST injury. The results of these studies will provide an assessment of the feasibility of using PRR recordings in patients with CST injury to control a prosthetic device.(full report)

Attentional Selection for Learning and Recognition of Objects in Cluttered Scenes
Ueli Rutishauser, Dirk Walther, Christof Koch, and Pietro Perona

The problem of serial processing of highly complex visual stimuli containing multiple objects is not only faced by humans and other primates, but also by machine vision systems. Advanced object recognition algorithms are capable of achieving very good recognition performance with objects learned from a single image (one-shot learning). These algorithms perform well as long as they are trained on images in which a major part of the image is occupied by the object to be learned and recognized. As soon as major parts of an image are occupied by clutter it becomes impossible to learn from such images without manual pre-labeling. These approaches are thus not suitable in an unsupervised environment, as they would mainly learn background clutter instead of the actual objects. (full report)

Perception of Mirror Surfaces
Silvio Savarese, Fei Fei Li, Pietro Perona

Abstract. The aim of our work is to investigate how the human visual system perceives specular surfaces and which cues can be used to recover the shape of such class of objects. Our experiments show that mirror reflections are a weak cue for most human observers when additional information is not available. (full report)

3D Reconstruction of Specular Surfaces
Silvio Savarese, Min Chen and Pietro Perona

Abstract. Specular reflections carry valuable information on surface shapes. A curved mirror surface produces "distorted" images of the surrounding world. For example a straight line reflected by a curved mirror is in general a curve. It is clear that such distortions are systematically related to the shape of the surface. Our goal is to explore the geometry linking the shape of a curved mirror surface to the distortions produced on a scene it reflects. To this effect, we assume a simple known (calibrated) scene composed of lines passing through a point. We demonstrate that local shape geometry of the surface may be recovered from local deformation of the reflected images of at least three intersecting lines. (full report)

Nano-to-Micro Self-Assembly Using Shear Flow Devices
Chi-Yuan Shih, Siyang Zheng, Ellis Meng, Yu-Chong Tai (Yi Liu and J. Frazer Stoddart)

It will be extremely useful if there’s a way to precisely assemble nano-materials into micro- or even meso-scale devices. For example, our long-term goal is to use massively architected motor-molecules [1] to build muscle-like actuators, in which these molecules work in parallel to output large forces. Unfortunately, the lack of such an assembly method is still the major barrier in the whole bottom-up nanotechnology field. This work aims at attacking this problem and as an important first step, we report here the successful development of a much improved shear-flow-enhanced self-assembly method over the baseline spontaneous assembly method in test tubes [2]. More specifically, we have engineered special thiolated model molecules (bisdisulfide/C28H34O4S4) and demonstrated the nano-to-micro self-assembly using thiol-gold bonding chemistry. Our method has produced gold/molecule aggregates as big as 50_m that are completely made of 30nm gold nanoparticles and 3nm model molecules. Fig.1 shows the idea of our shear-flow assembly. The interface of two shear flows is where gold nanoparticles meet with the thiolated molecules, herein the aggregation happens. The important advantages of this approach are twofold. The first is to limit the assembly only at interface for controlled assembly. The second advantage is the unsaturated growth of aggregate because shear flows continue to supply fresh nano-materials to the interface, leading to large aggregates. To implement this design, we fabricate two types of shear flow devices (Fig.2). For water or ethanol solvent system, PDMS/glass devices are used for easy plumbing and observation. For non-polar solvents like acetone and dichloromethane, glass/silicon devices are used to avoid PDMS swelling. (full report)

A Biosphere for Studying Neural Circuits of Drosophila melanogaster
Jasper Simon, Michael Dickinson

Research Proposal. Observation rather than experimentation dominates the study of animal behavior, a limit to our understanding. We require the ability to study behavior while aspects of an animal's environment can be controlled. To meet this goal, I plan a biosphere in which I can control various parameters to recapitulate the pertinent aspects of an animal's natural environment.

Seasonal change and undesirable habitats force animals to assess local resources and decide between to stay or to move somewhere potentially more desirable. Cues from both the environment and an animal's current internal state influence such decisions. What mechanisms underlie the ability to integrate and process these cues? Is movement directed simply by cue saliency? Or do animals carry out some rudimentary cost-benefit analysis?

Within a neuroethological context, resource leaving in the fruit fly Drosophila melanogaster provides a useful model to study such elementary decision making. With the molecular tools available in Drosophila, I propose to study the neural circuits involved in this process.

Neurogenetic Dissection of Resource Choice in the Fruit Fly Drosophila melanogaster
Jasper Simon, Michael Dickinson

Abstract. I propose to study the neurogenetic mechanisms that underlie resource choice in the fruit fly Drosophila melanogaster. Specifically, how do genes regulate the decision to leave resources? In natural environments the distribution and abundance of resources vary over space and time—quite scarce during certain times in the life of a fly. Thus, it seems flies would stay indefinitely on an established resource, but casual observation proves this false. At various times scales: moment-to-moment, over the course of a day, or throughout a lifetime, flies leave resources. What external and internal cues influence the probability to leave? How do these cues interact? Moreover, this behavior initiates dispersal and has implications for the animal’s life history. Within a neuroethologcal context, resource leaving in flies provides a useful model to study elementary decision-making in a simple nervous system. I aim to characterize, identify, and define the relative contribution of external sensory cues, internal state cues, and their interactions in the determination of resource choice. Using molecular and population genetic approaches, I will attempt to identify the neuronal circuits and genes that participate in the regulation of resource choice.

Nonlinear Femtosecond Pulse Delivery in Optical Fibers
Mankei Tsang, Demetri Psaltis

Abstract. We investigate two methods to compensate for dispersion and nonlinearity in optical fiber ultrashort pulse propagation for applications in biomedical imaging and optical communications. One method makes use of numerical reverse propagation results to preshape an input optical pulse, such that an output pulse of any shape, width and intensity can be produced amidst all the linear and nonlinear distortions. Another method uses midway spectral phase conjugation to compensate for all dispersion and most nonlinearity. (full report)

Trace and Delay Fear Conditioning and its Dependence on Awareness in Humans
Tsuchia, N., Koch, C

Previous studies of associative learning implicate higher-level cognitive processes in some forms of classical conditioning. An ongoing debate is concerned with the extent to which attention and awareness are necessary for trace but not delay eye blink conditioning (Clark, R.E. & Squire, L.R. (1998) Science 280, 77-81; Lovibond, P.F. & Shanks, D. (2002) J. Exp. Psychol. Anim. Behav. Process 28, 38-42]. In trace conditioning, a short interval is interposed between the termination of the conditioned stimulus CS and the onset of the unconditioned stimulus US. In delay conditioning, the CS and US overlap. We investigate the extent to which human classical fear conditioning depends on working memory and attention. (full report)

Averaging Methods for Control of Biomimetic Locomotion
Patricio Vela, Joel Burdick

Abstract. Biomimetic control systems are inherently difficult to control because they typically fall within the class of nonlinear control systems known as underactuated systems with drift (a few models do exist without drift). The nature of underactuated systems is such that smooth exponentially stabilizing feedback laws do not exist, and therefore require time-varying or non-smooth (either continuous or discontinuous) feedback strategies. Since many biomimetic systems exhibit time-periodic behavior, it is advantageous to examine the role of time-periodic forcing elements on the underlying equations of motion.

Current work focuses on developing a general averaging theory that works to arbitrary orders of approximation so as to better understand the nonlinear response of dynamical systems. From this, it should be possible to develop feedback control laws that are exponentially stabilizing. Recent work has developed the averaging methods and applied them to various biomimetic systems, such as a carangiform fish, a snakeboard, and a kinematic biped. The nature of the averaging results and subsequent control laws demonstrate that the approach may hold generally, irrespective of the locomotive strategy, i.e., discontinuous legged locomotion, land locomotion, or hydrodynamic locomotion. (full report)

Automated Event Detection in Underwater Video
Dirk Walther, Duane Edgington, Karen A. Salamy, Michael Risi, R. E. Sherlock, and Christof Koch

Remotely operated underwater vehicles (ROVs) become increasingly important as a tool for obtaining quantitative data on the distribution and abundance of oceanic animals. Using video cameras, it is possible to make quantitative video transects (QVT) through the water, providing high-resolution data at the scale of the individual animals and their natural aggregation patterns. The current manual method of analyzing QVT video by trained scientists is very labor intensive and poses a serious limitation to the amount of data that can be obtained from ROV dives. (full report)

Evolutionary Design Synthesis – From Sensors to Controllers
Yizhen Zhang, Alcherio Martinoli, Erik Antonsson
Collaborators: Jonathan Litt, Edmond Wong (NASA Glenn Center

Abstract. In this project, an automated engineering design synthesis methodology based on evolutionary methodology is being explored, with special interest on design and optimization of distributed embodied systems. Two case studies have been considered so far; the first one concerns the design of a collective sensory system for traffic monitoring purposes, while the second one deals with the development of neural-based robot controllers for turbine blades inspection. It has been shown that the evolutionary methodology is able to address the engineering design challenges present in the case studies as well as other complex design problems, and synthesize novel design solutions of good quality. Moreover, the fitness function can be formulated as an aggregation of fuzzy design preferences with different weights and trade-off strategies leading to an automatic generation of the complete Pareto-optimal frontier.. (full report)