Register Submit Abstract Contacts Directions Program Home

15th Joint Symposium on Neural Computation

University of California, Irvine
Irvine, California

Social Sciences Plaza, Bldg A, Rm 1100



May 31, 2008



In 1994, the Institute for Neural Computation at UCSD hosted the first Joint Symposium on Neural Computation with Caltech. This Symposium brought together students and faculty for a day of short presentations. Since then, this Symposium has rotated between San Diego, Caltech, UCI, UCLA and USC.
Past Programs Program Committee
PROGRAM
8:15am Registration and Coffee
9:00am

David Kleinfeld, UCSD
Department of Physics and Graduate Program in Neurosciences

Perception and Action:  How the Vibrissa System Encodes Relative Object Location


Sensory perception in natural environments involves the dual challenge to encode external stimuli and manage the influence of changes in body position that alter the sensory field.  I will discuss a mechanism used to integrate sensory signals elicited by both external stimuli and motor activity in the context of trained rats that perform an active sensory task.  This integration enables the rodent to estimate object position in a head-centered reference frame.  More generally, our results delineate a computation that is likely to occur in all active sensorimotor systems.

9:30am    

Jeff Krichmar, UCI
Department of Cognitive Sciences

Neuromodulation and Time-Dependent Plasticity in a Model of Foraging Behavior


In foraging behavior, where an animal searches for food caches, it is imperative for the animal to remember the locations and routes to these caches. An important consideration is the means by which the organism takes the appropriate actions to lead it to a goal that satisfies a particular need. We introduce a time-dependent plasticity rule that biases movement in a particular direction by developing asymmetric neuronal receptive fields through experience. The model contains hippocampal areas that respond differentially to locations in space, frontal cortex areas that respond to different salient cues from the environment, and neuromodulators that respond to rewards and costs. This model suggests a means by which neuromodulated time-dependent plasticity in the frontal cortex can facilitate action selection.  It also suggests how these neuronal responses may lead to successful performance in a foraging task.

10:00am

Gert Cauwenberghs, UCSD
Department of Biology

Scalable Neuromorphic Cortical Systems


The quest to build machines that think and act like humans is impeded by the massive complexity of the human brain and by our limited knowledge of how the brain functions. Despite significant advances towards naturally intelligent computing using neuromorphic engineering approaches to computer architecture, the majority of electronic neural systems existing to date exhibit primitive function and serve as a proof of concept in modeling isolated parts of the brain and the nervous system. To scale up the functionality of these systems towards brain-like computing, several researchers have adopted an event-based spiking representation to interface multichip neuromorphic processors, sensors, and actuators. I will present our work on scalable neural architecture with reconfigurable connectivity and dynamic synaptic plasticity, which extends to large systems approaching the computational bandwidth and efficiency of mammalian cortex, implemented using custom designed silicon microchips.  We anticipate further advances in density and energy efficiency by migrating, using similar system-level principles, to nanotechnology in massively parallel distributed architecture. I will also address challenges in configuring and training the hardware and highlight some promising approaches that employ hierarchical organization and the ubiquitous availability of human-assisted training data over the internet.

10:30am

Michael Arbib, USC
Fletcher Jones Professor of Computer Science, Department of Biological Sciences, Biomedical Engineering, Electrical Engineering, Neuroscience and Psychology; Director, USC Brain Project


Describing Visual Scenes: Towards a Neurolinguistics Based on Construction Grammar


The present talkis part of a larger effort to locate the production and perception of  language within the broader context of brain mechanisms for action and perception more generally. Here we model function in terms of the competition and cooperation of schemas. We use the task of describing visual scenes to explore the suitability of Construction Grammar as an appropriate framework for a schema-based linguistics. We recall the early VISIONS model of schema-based computer analysis of static visual scenes and then introduce SemRep as a graphical representation of dynamic visual scenes designed to support the generation of varied descriptions of episodes. We report preliminary results on implementing the production of sentences using Template Construction Grammar (TCG), a new form of construction grammar distinguished by its use of SemRep to express semantics. We summarize data on neural correlates relevant to future work on TCG within the context of neurolinguistics, and show how the relation between SemRep and TCG can serve as the basis for modeling language comprehension. (Work with JinYong Lee)

11:00am    Carlos Brody, Princeton
Department of Molecular Biology

Simple Models for Complex Data


Very often, models are hand-designed to reproduce aspects of experimental data that the modelers have deemed "important."  But often data is complex, with strong heterogeneity across different neurons, and has many more features than those in the model. How can we capture all those features in a meaningful way? We use a simple numerical method to fit a network of model neurons to reproduce all firing rate traces from an experiment. We are using these models to explore the role of heterogeneity in neural computation.
11:30am Poster Highlights
12:00pm

Lunch Break & Poster Session

2:00pm

Zhong-Lin Lu, USC
William M. Keck Chair in Cognitive Neuroscience, Department of Psychology and Biomedical Engineering, Dana and David Dornsife Cognitive Neuroscience Imaging Center

The Quick Methods: Bayesian Adaptive Estimation of Psychological Functions

Authors: Zhong-Lin Lu, Luis Lesmes, Jongsoo Baek, Simon Jeon, Barbara A. Dosher, & Thomas Albright

Adaptive methods are well-known in psychophysics.  The idea is to use dynamic stimulus placement strategies based on subject’s responses to optimize the efficiency of data collection.  Development of adaptive methods has mostly focused on estimating properties of psychometric functions.  Based on subject’s responses, these methods target stimuli to pre-specified regions of the empirical psychometric functions (e.g. threshold region). Our goal is to develop and test adaptive methods for characterizing other psychological functions.  Here, we review some recent progress in applying the general framework to measurements of TvC functions, d' psychometric functions, contrast sensitivity functions, and forgetting functions.

2:30pm

Barbara Dosher, UCI
Department of Cognitive Sciences; Dean, School of Social Sciences

Mechanisms and Models of Perceptual State Changes


Many human systems exhibit performance changes in responses to changes in observer state, such as transient attention and perceptual learning. Changes in state can often be characterized within the context of functional models of the observer. Assays, for example, of attentional or perceptual learning effects in noisy or noiseless environments identify the mechanisms of perceptual state change as a change in the ability to exclude external noise or distractors through retuning, as stimulus enhancement through increased response, or as a change in gain control properties of the visual system. Each mechanism has a characteristic signature in performance, which in turn may illuminate the nature of change in processing.

3:00pm

John Reynolds, Salk Institute
Systems Neurobiology Laboratory


Mapping the Microcircuitry of Attention: Attentional Modulation Varies Across Cell Classes in Visual Area V4

Authors:  John Reynolds & Jude F. Mitchell

Cortical neurons differ from one another in important ways, including their neurochemical properties, patterns of connectivity, laminar distribution, gene expression patterns and developmental origin. Previous studies of attention have not sought to distinguish among different classes of neurons. We therefore know almost nothing about the circuitry that transforms attentional feedback signals into improved visual processing. Studies in the slice and in anesthetized animals find that parvalbumin expressing GABA-ergic interneurons with the morphologies of basket and chandelier cells have short duration action potentials, whereas most excitatory cell classes have longer duration action potentials, a difference that is due to expression of different classes of sodium and potassium channels. We thus examined differences in attentional modulation across visual area V4 neurons classified on the basis of action potential width. The distribution of action potential widths in area V4 is clearly bimodal. We find substantial differences in the basic response properties of these two classes of neurons, including their baseline firing rates, the strength of their stimulus-evoked responses, as well as qualitative differences in the types of variability of the neuronal response across classes. We also find qualitative differences in how the two neuronal classes are modulated by attention, including differences in how attention modulates firing rate and differences in the attention-dependent reduction in response variability among the two classes of neurons. Narrow spiking neurons show a marked low frequency fluctuation in firing rate that is diminished by attention. Many broad spiking neurons show burstiness in their responses that is diminished by attention. The discovery of differences in attentional modulation of firing rate and neuronal noise represents a key step forward in developing circuit-level models of attention and visual processing.

3:30pm

Richard Andersen, Caltech
James G. Boswell Professor of Neuroscience

"Free Choice" Activates Decision Circuits


How we choose between alternatives is a very interesting subject of research in the brain sciences. However, the neural circuits involved are not well understood. I will describe two recent experiments from our lab using "free choice." The first shows that the posterior parietal cortex, a cortical area often considered largely sensory in nature, encodes freely selected motor plans.  The second experiment involved simultaneous recordings between posterior parietal cortex and dorsal premotor cortex. We found greater correlations of spikes in one area with local field potentials in the other when the subjects were making free choices compared to following instructions. Importantly, only a sub-population of cells in these two areas showed significant correlation and these neurons are the first to encode the decision.  Thus they may represent a specialized circuit that coordinates activity between cortical areas during decision-making. 

4:00pm

Stan Schein, M.D., Ph.D. UCLA
Professor, Psychology - Behavioral Neuroscience, California NanoSystems Institute, Brain Research Institute


How Small Numbers of Quanta of a Neurotransmitter Can Encode Small Changes in the Receptor Potential of Extraordinarily Sensitive Receptor Cells Like Rod Photoreceptors and Hair Cells


Extraordinarily sensitive sensory systems face two problems.  First, the receptor cell must be able to transduce an extremely weak stimulus like a single photon in the case of a rod photoreceptor. Second, the receptor cell must be able to transmit its small electrical signal to its target neurons by a small change in the number of packets of neurotransmitter ("quanta") that it releases. The quantal and stochastic nature of neurotransmitter release, along with the small numbers involved, impose inescapable mathematical constraints on efficient transmission of such signals.

For example, even though the receptor potential for absorption of one photon is just one millivolt, a mammalian rod photoreceptor can transmit its signal with 25%-50% efficiency. A hair cell in the auditory or vestibular system can transmit its signal for movement of the tips of its stereocilia by a nanometer, a displacement that produces a receptor potential considerably less than one millivolt. And, nonspiking electroreceptors can transmit voltage signals in the range of microvolts. These tiny changes in membrane potential produce tiny changes in the rate of quantal release. Moreover, even though these receptor cells may release quanta continuously at high rates, the postsynaptic target cell has only a limited window of time in which to count quanta. As a result, the difference in the target cell's quantal count between the unstimulated condition and the stimulated condition is very small.

In light of the small difference in quantal count, quantal noise would create a serious problem. For example, if transmission of a single-photon event by a mammalian rod employed random (Poisson) quantal release, the rod bipolar dendrite might compare a Poisson distribution of quantal count ND = 10 ±  quanta (mean count ± SD) in the Dark with NL = 8 ± quanta for one photon (in the "Light"). These two count distributions overlap extensively. To reduce overlap we suggested that quantal release must be regular ("clockwork") rather than random, regularity like that of a ~64th order Erlang process. In other words, each quantal release event would occur after r ≈ 64 Poisson events have been counted, reducing standard deviations (quantal noise) by~8-fold compared to a random (Poisson) quantal event stream.

Here we add a general concept: Some number N of regular events (e.g., quanta or spikes) is equivalent to rN Poisson events, where r is the Erlang order in units of Poisson events per quantum. A modification of Signal Detection Theory can then use this number of Poisson events to determine discriminability between two conditions, like Dark and Light or Sound and Silence.
4:30pm     Bosco Tjan, USC
Department of Psychology and Neuroscience Graduate Program

Crowding in Peripheral Vision

Form vision is often associated with the high-resolution central vision provided by the fovea.  However, patients who lose their central vision from diseases such as macular degeneration must rely on their peripheral visual fields to recognize objects, identify faces, and read.  Compared to the fovea, the periphery is far less capable of these types of form vision, even after its poor spatial resolution has been compensated by magnification and contrast enhancement.  For example, reading in the periphery is laboriously slow, and objects often cannot be identified in a cluttered scene.  We study why peripheral form vision is qualitatively inferior to central vision.  Results from our recent psychophysical and neuroimaging experiments suggest that form vision in the periphery is hindered by a lost of the relative positional information between visual features at an early stage of visual processing, probably V1 or V2; as a result, the subsequent stages have to infer object identity from bags of unorganized features.
Register Submit Abstract Contacts Directions Program Home

Site maintained by: jsnc@caltech.edu
Last update: 5/6/08