* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Affiliates Day Poster Joseph Young
Survey
Document related concepts
Catastrophic interference wikipedia , lookup
Nonsynaptic plasticity wikipedia , lookup
Synaptic gating wikipedia , lookup
Types of artificial neural networks wikipedia , lookup
Feature detection (nervous system) wikipedia , lookup
Neural coding wikipedia , lookup
Single-unit recording wikipedia , lookup
Donald O. Hebb wikipedia , lookup
Metastability in the brain wikipedia , lookup
Process tracing wikipedia , lookup
Activity-dependent plasticity wikipedia , lookup
Nervous system network models wikipedia , lookup
Eyeblink conditioning wikipedia , lookup
Psychological behaviorism wikipedia , lookup
Transcript
Causality Analysis of Neuron Spike Timings Recorded During Task Learning Joseph Young1, Valentin Dragoi2, Behnaam Aazhang1 1 Department of Electrical and Computer Engineering, Rice University, Houston, USA 2 Department of Neurobiology and Anatomy, The University of Texas Medical School at Houston, Houston, USA Understanding Learning Experimental Setup - Visual Task Learning: Well observed phenomenon • “Practice makes perfect” • Poorly understood in terms of neurophysiology Directed Information Estimation Rhesus Monkeys Performed Rotation Recognition Task: Definition Time 2 3 4 5 Fixation Target … Test Decision 500 ms 300 ms 1000 ms Rotated or not? 300 ms Low f Spike Channels Neuronal recordings were taken from Rhesus monkeys’ visual cortex (V4) while they performed a visual task Field Potential Channels MAP Multichannel Acquisition Processor (Plexon) Contact Joseph Young [email protected] http://www.owlnet.rice.edu/~jy46 Department of Electrical and Computer Engineering Rice University, Houston, TX Single-units amplified & filtered Threshold i=1 H(Yi |Y i 1 Spike-Field Coherence (SFC) Was Higher During Learning • SFC measures synchronization between single neuron and local population as a function of frequency • SFC was greatest in theta band during learning • After learning, changes in SFC disappeared for all frequencies , X i) • • … Test 200 ms 200 ms Decision Analysis will focus on 200 ms window after each image onset to avoid synchronicity due to stimulus presentation 21 neurons to analyze from 17 channels (electrodes) Data from 623 trials of task Contributions Causality Analysis Will: • Quantify influence between neurons • Provide insight into how influence changes during learning • Suggest procedure / areas for stimulated learning ; 11 • Target { Neuronal Synchronization During Task Learning Moving From Synchronization to Causation High f n X Neural Spike Analysis For the decision period (5), the monkey must release a bar by the end of the period if the test image is a rotated version of the target image. Correct decision results in juice for the monkey. Universal Estimation of Directed Information • Feed sequences of neuronal spikes into the context-tree weighting (CTW) algorithm CTW • M-ary tree with weights • Provides probability estimates to plug into directed information equations (quantifying causality) Procedure: • Slide M-length window across an input sequence of finitealphabet size • Match window with context of a bottom level node (e.g. “01”) and update parameters based on next symbol after the window H(Y n ||X n ) 1500 ms Neuronal Data Collection Neuronal Signals from V4 ) Estimation • Plug in probabilities from Context-Tree • Estimate DI flow between neuron pairs Fixation New Tech Allows Ensemble Recording • Can observe neuronal network • Functional / effective connectivity analysis • Research at UTHSC studied neuronal synchronization in Rhesus monkey brains during task learning • We will further analyze this same data set using directed information (DI) between neurons to determine causality 1 Causally Conditioned Entropy: H(Y n ||X n ) , Levels of Analysis 1990s Focus on Single Neurons • Learning thought to involve modifications in the brain on the order of single cells and single synapses • No support - network more promising = H(Y n ) 1 [blank] i=1 I(X i ; Yi |Y i { Objective: Discover how neuronal population connectivity alters over the time course of learning • Quantify causality between recorded spike timings • Produce graphical model of connectivity Preamplifier I(X n ! Y n ) , n X 1 0 01 10 00 Future Work: • Carry out recordings on larger ensembles to account for effective connectivity between different areas during learning • Stimulate population areas to improve learning • Develop model that describes the time course of learning in the visual cortex Example context-tree for M=2 Spikes References 1. Ye Wang and Valentin Dragoi. Rapid learning in visual cortical networks. eLife, 4:e08417, aug 2015. 2. J. Jiao, H. H. Permuter, L. Zhao, Y. H. Kim, and T. Weissman. Universal estimation of directed information. IEEE Transactions on Information Theory, 59(10):6220–6242, Oct 2013. 3. F. M. J. Willems, Y. M. Shtarkov, and T. J. Tjalkens. The context-tree weighting method: basic properties. IEEE Transactions on Information Theory, 41(3):653–664, May 1995. 4. 5. 6. Sagi, D., & Tanne, D. Perceptual learning: learning to see. Current opinion in neurobiology. 1994. Karni, A., & Bertini, G. Learning perceptual skills: behavioral probes into adult cortical plasticity. Current opinion in neurobiology. 1997. Rhesus image: From Einar Fredriksen (http://flickr.com/photos/85198067@N00/4185543087), seen at https://commons.wikimedia.org/wiki/ File:Macaca_mulatta_in_Guiyang.jpg