Cosyne 2008 Workshops
March 3-4, 2008
Snow Bird, Utah
Workshop Title
Real-time processing and the processing of time.
Organizer(s)
Sophie Deneve (ECS): sophie.deneve@ens.fr
Dean Buonomano (UCLA): dbuono@ucla.edu
Abstract
The brain's ability to make sense of the "blooming buzzing confusion" that is our sensory environment relies on the ability of the cortex to extract patterns out of a continuous stream of sensory information. This information is encoded in both space (which afferents are active) and time (the temporal patterns with which afferents are activated). The processing of spatio-temporal patterns must be performed continuously in a real-time fashion. The brain needs to extract information about timing, but also, to use time in order to process information.
Using time to process information: We constantly receive inputs from our sensory receptors and this sensory information needs to be integrated over time. However, in most cases, this integration cannot rely on simple temporal summation of the sensory input: This is because the world and the state of the variables that one tries to infer is likely to vary unpredictably at any time: While it is important to extract as much information as possible from the recent input, it is equally important to forget older inputs that are no longer relevant. Traditional approaches to the most studied perceptual system, the visual system, have emphasized static problems, such as how to localize or recognize an object, or segment an image. This is very far from natural vision whose aim is to detect a succession of visual events, as far as possible, on-line.
Extracting information about timing: Natural stimuli, such as speech, are rich in spatial as well as temporal information. Indeed, speech, for example, can be essentially collapsed into a purely temporal code - Morse Code - which can be decoded in either the auditory, visual or somatosensory modalities.
A neglect of the temporal dimension can be observed in many computational models of neural coding. Neurons fire spikes, which is inherently a time-based code. However, rate coding approach assumes that information is contained in the mean number of action potential. This deprives the spikes of their "event based" nature and relegates them as random samples of the only relevant signal: the rate. Similarly, while significant advances have been made regarding how cortical circuits discriminate static spatial patterns, relatively little is known about how the brain decodes information about timing and temporal patterns.
In this workshop we will bring together neurophysiologist and computational neuroscientists that have worked on 1. Sensory processing on-line, in real time. 2. Neural processing and representation of temporal information
We will in particular investigate whether these two related topics face the same issues and are based on similar neural mechanisms. In particular, do spike times or burst of spikes precisely signal relevant events? Is it related to predictive coding? What are the contributions of inhibition, synaptic short term plasticity, adaptation, refractoriness for perception on-line and perception of time? How do networks of neurons decode temporal information? Is it useful to formalize these problems in a statistical framework, for example as a hidden Markov Model?
Speakers
| Dan Butts | How does natural vision drive precise timing in the visual system, and why? |
| Sophie Deneve | Spiking neural networks performing Bayesian inference over time. |
| Michael Kilgard | Effect of Neural Correlations on Speech Discrimination |
| Michael Mauk | Timing and Temporal Coding in the Cerebellum |
| Gilad Silberberg | Temporal Properties of Excitation & Inhibition in Neocortical Microcircuits |
| Fritz Sommer | title tbd |
| Haim Sompolinsky | Processing of Time Through Space |
| Misha Tsodyks | title tbd |
| Dean Buonomano | Telling Time with State-Dependent Networks and Short-Term Synaptic Plasticity |
