{[ promptMessage ]}

Bookmark it

{[ promptMessage ]}

REP-2009-473 - TASK-BASED SPIKETRAIN DISCRIMINATION AND AN...

Info iconThis preview shows pages 1–7. Sign up to view the full content.

View Full Document Right Arrow Icon
TASK-BASED SPIKETRAIN DISCRIMINATION AND AN APPLICATION TO NEURAL REINFORCEMENT By NATHAN D. VANDERKRAATS A PROPOSAL PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 2008 1
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
c circlecopyrt 2008 Nathan D. VanderKraats 2
Background image of page 2
TABLE OF CONTENTS page LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6 ABSTRACT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8 CHAPTER 1 BACKGROUND MATERIAL . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2 Neuroscience Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.1 Basic neural dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 9 1.2.2 Functional Brain Regions . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2.3 Neural Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12 1.2.4 Spike Timing Dependent Plasticity . . . . . . . . . . . . . . . . . . 15 1.3 Auditory System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 16 1.4 Meddis Inner-Hair Cell Model . . . . . . . . . . . . . . . . . . . . . . . . . 18 1.5 Machine Learning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19 1.5.1 Support Vector Machines . . . . . . . . . . . . . . . . . . . . . . . . 20 1.5.2 Perceptron . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 2 ON THE MUTUAL INFORMATION BETWEEN TASK-BASED STIMULI AND SPIKETRAIN RESPONSES FOR LARGE NEURAL POPULATIONS . . 22 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.1.1 Informational Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.1.2 Task-Based Reconstruction . . . . . . . . . . . . . . . . . . . . . . . 23 2.1.3 A Tighter Information-Theoretic Bound . . . . . . . . . . . . . . . . 25 2.1.4 Previous Information-Theoretic Results . . . . . . . . . . . . . . . . 26 2.2 Description of Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2.2 Auditory input . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27 2.2.3 Spiking Neuron Model . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.2.4 Decoder Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30 2.2.5 Feature Space Candidates . . . . . . . . . . . . . . . . . . . . . . . 31 2.2.6 Statistical Classifier . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 2.3 A Lower Bound on MI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 3 DECODING METHOD RESULTS . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.1 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.1.1 Description of Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . 34 3.1.2 Baseline Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38 3.1.3 Single-Layer Networks . . . . . . . . . . . . . . . . . . . . . . . . . 40 3
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
3.1.4 Two-Layer Networks . . . . . . . . . . . . . . . . . . . . . . . . . . 43 3.2 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43 4 TRAINING AN STDP-ENABLED NEURON WITH AN INNOCUOUS TEACHING SIGNAL . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46 4.1.1 A Reinforcement Learning Model . . . . . . . . . . . . . . . . . . . 47 4.2 Model of the Neuron and Inputs . . . . . . . . . . . . . . . . . . . . . . . . 48 4.3 Definition of the Learning Task . . . . . . . . . . . . . . . . . . . . . . . . 48 4.3.1 Experimental Overview . . . . . . . . . . . . . . . . . . . . . . . . . 48 4.3.2 Learning Task . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 4.3.3 Innocuous Reinforcement . . . . . . . . . . . . . . . . . . . . . . . . 51 4.4 Teaching Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.4.1 Objective Function . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 4.4.2 Optimal Teaching Perturbations . . . . . . . . . . . . . . . . . . . . 54 4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4.5.1 Classification Improvement . . . . . . . . . . . . . . . . . . . . . . . 58 4.5.2 Innocuousness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60 5 FUTURE WORK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 5.1 Decoder Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61 5.1.1 Mutual Information Bound Given Classification Accuracy . . . . . . 61 5.1.2 Decoder Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . 62 5.2 Global Neural Learning Method Extensions . . . . . . . . . . . . . . . . . 62 5.2.1 Model Generalizations . . . . . . . . . . . . . . . . . . . . . . . . . 62 5.2.2 Alternate Tasks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63 5.2.3 Incremental Learning Algorithm . . . . . . . . . . . . . . . . . . . . 63 APPENDIX: MEDDIS INNER-HAIR CELL MODEL PARAMETERS . . . . . . . . 64 REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66 4
Background image of page 4
LIST OF TABLES Table page 3-1 Baseline results on the auditory nerve, sorted by task. Response windows are a pair (offset, window size) denoting the offset and length of the response window in ms . The optimal gain field concerns the sound’s initial gain for the Meddis model. The classification accuracies are obtained using a spike times representation and the five-way cross validation described in the text over 100,000 samples. . . 39 3-2 Results for the single-layer networks, sorted by network type. . . . . . . . . . . . 41 3-3 Results for the two-layer networks, sorted by network type. . . . . . . . . . . . . 44 A-1 Meddis model parameters for 40 center frequency groups. . . . . . . . . . . . . . 65 5
Background image of page 5

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full Document Right Arrow Icon
LIST OF FIGURES Figure page 1-1 Simplified drawing of a neuron. . . . . . . . . . . . . . . . . . . . . . . . . . . . 10 1-2 Basic neural dynamics for a single neuron with two inputs. . . . . . . . . . . . 10 1-3 STDP synaptic weight update as a function of time between the presynaptic and postsynaptic spikes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15 2-1 Two ways to use the classification information from a discriminant.
Background image of page 6
Image of page 7
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}