Files in this item



application/pdfWANG-DISSERTATION-2018.pdf (5MB)
(no description provided)PDF


Title:Polychronization as a mechanism for language acquisition in spiking neural networks
Author(s):Wang, Felix Y
Director of Research:Levinson, Stephen E.
Doctoral Committee Chair(s):Levinson, Stephen E.
Doctoral Committee Member(s):Hasegawa-Johnson, Mark; Rothganger, Fred; Smaragdis, Paris
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Language Acquisition
Spiking Neural Network
Neural Computing
Spiking Simulation
Associative Memory
Pattern Recognition
Supervised Learning
Multi-modal Learning
Abstract:The capacity of an intelligent agent to process complex patterns in signals such as language rests heavily on the nature of the internal representation of the relevant information. Furthermore, the acquisition of internal representation is an inherently closed-loop process in which an intelligent agent enters into a conversation with its environment. The result is the construction of a necessarily generative model of language, where semantics are grounded in an agent's sensory-motor experience by way of an associative memory. This work explores the mechanisms underlying language acquisition by investigating the function and architecture of the neocortex, with the ultimate goal of understanding how mental states might arise from spiking activity. In particular, we focus on the phenomenon of polychronization, which may be described as the self-organization of a spiking neural network as a result of the interplay between network structure, spiking activity, and synaptic plasticity. What emerges are groups of neurons exhibiting time-locked patterns of spiking, reproducible spatio-temporal stamps consisting of the precisely timed activations of their constituent neurons. At a high level, these polychronous neural groups may be thought of as a form of temporal encoding of information within the network. We propose that this representation is well suited to language acquisition, as it naturally resembles the spatio-temporal patterns found in the speech signal, as well as other real-world signals. Toward this end, we develop a supervised method for training a spiking neural network to learn and recognize patterns that are relevant to language, such as those corresponding to phonetic primitives. We show that even for a simplified model, the mechanism of polychronization is capable of processing and representing such patterns, providing a basis for language acquisition in spiking neural networks.
Issue Date:2018-06-21
Rights Information:Copyright 2018 Felix Y. Wang
Date Available in IDEALS:2018-09-27
Date Deposited:2018-08

This item appears in the following Collection(s)

Item Statistics