Time underlies many interesting human behaviors. Thus, the question of how to represent time in connectionist models is very important. One approach is to represent time implicitly by its effects on processing rather than explicitly (as in a spatial representation). The current report develops a proposal along these lines first described by Jordan (1986) which involves the use of recurrent links in order to provide networks with a dynamic memory. In this approach, hidden unit patterns are fed back to themselves; the internal representations which develop thus reflect task demands in the context of prior internal states. A set of simulations is reported which range from relatively simple problems (temporal version of XOR) to discovering syntactic/semantic features for words. The networks are able to learn interesting internal representations which incorporate task demands with memory demands; indeed, in this approach the notion of memory is inextricably bound up with task processing. These representations reveal a rich structure, which allows them to be highly context-dependent while also expressing generalizations across classes of items. These representations suggest a method for representing lexical categories and the type/token distinction.

A major goal of research on networks of neuron-like processing units is to discover efficient learning procedures that allow these networks to construct complex internal representations of their environment. The learning procedures must be capable of modifying the connection strengths in such a way that internal units which are not part of the input or output come to represent important features of the task domain. Several interesting gradient-descent procedures have recently been discovered. Each connection computes the derivative, with respect to the connection strength, of a global measure of the error in the performance of the network. The strength is then adjusted in the direction that decreases the error. These relatively simple, gradient-descent learning procedures work well for small tasks and the new challenge is to find ways of improving their convergence rate and their generalization abilities so that they can be applied to larger, more realistic tasks.

I present a general taxonomy of neural net architectures for processing time-varying patterns. This taxonomy subsumes many existing architectures in the literature, and points to several promising architectures that have yet to be examined. Any architecture that processes timevarying patterns requires two conceptually distinct components: a short-term memory that holds on to relevant past events and an associator that uses the short-term memory to classify or predict. My taxonomy is based on a characterization of short-term memory models along the dimensions of form, content, and adaptability. Experiments on predicting future values of a financial time series (US dollar--Swiss franc exchange rates) are presented using several alternative memory models. The results of these experiments serve as a baseline against which more sophisticated architectures can be compared. Neural networks have proven to be a promising alternative to traditional techniques for nonlinear temporal prediction t...

In this paper we present a Lagrange-multiplier formulation of discrete constrained optimization problems, the associated discrete-space first-order necessary and sufficient conditions for saddle points, and an efficient first-order search procedure that looks for saddle points in discrete space. Our new theory provides a strong mathematical foundation for solving general nonlinear discrete optimization problems. Specifically, we propose a new vector-based definition of descent directions in discrete space and show that the new definition does not obey the rules of calculus in continuous space. Starting from the concept of saddle points and using only vector calculus, we then prove the discrete-space first-order necessary and sufficient conditions for saddle points. Using well-defined transformations on the constraint functions, we further prove that the set of discrete-space saddle points is the same as the set of constrained local minima, leading to the first-order necessary and sufficient conditions for constrained local minima. Based on the first-order conditions, we propose a local-search method to look for saddle points that satisfy the first-order conditions.

. Spiking neurons, receiving temporally encoded inputs, can compute radial basis functions (RBFs) by storing the relevant information in their delays. In this paper we show how these delays can be learned using exclusively locally available information (basically the time difference between the pre- and postsynaptic spike). Our approach gives rise to a biologically plausible algorithm for finding clusters in a high dimensional input space with networks of spiking neurons, even if the environment is changing dynamically. Furthermore, we show that our learning mechanism makes it possible that such RBF neurons can perform some kind of feature extraction where they recognize that only certain input coordinates carry relevant information. Finally we demonstrate that this model allows the recognition of temporal sequences even if they are distorted in various ways. 1. Introduction Radial basis functions (RBFs) have turned out to be among the most powerful artificial neural network types, e....

Auditory scene analysis is critical for complex auditory processing. We study auditory segregation from the neural network perspective, and develop a framework for primitive auditory scene analysis. The architecture is a laterally coupled two-dimensional network of relaxation oscillators with a global inhibitor. One dimension represents time and another one represents frequency. We show that this architecture, plus systematic delay lines, can in real time group auditory features into a stream by phase synchrony and segregate different streams by desynchronization. The network demonstrates a set of psychological phenomena regarding primitive auditory scene analysis, including dependency on frequency proximity and the rate of presentation, sequential capturing, and competition among different perceptual organizations. We offer a neurocomputational theory - shifting synchronization theory - for explaining how auditory segregation might be achieved in the brain, and the psychological pheno...

This thesis examines how artificial neural networks can benefit a large vocabulary, speaker independent, continuous speech recognition system. Currently, most speech recognition systems are based on hidden Markov models (HMMs), a statistical framework that supports both acoustic and temporal modeling. Despite their state-of-the-art performance, HMMs make a number of suboptimal modeling assumptions that limit their potential effectiveness. Neural networks avoid many of these assumptions, while they can also learn complex functions, generalize effectively, tolerate noise, and support parallelism. While neural networks can readily be applied to acoustic modeling, it is not yet clear how they can be used for temporal modeling. Therefore, we explore a class of systems called NN-HMM hybrids, in which neural networks perform acoustic modeling, and HMMs perform temporal modeling. We argue that a NN-HMM hybrid has several theoretical advantages over a pure HMM system, including better acoustic ...

Odors evoke synchronized oscillations and slow temporal patterns in antennal lobe neurons and fast oscillations in the mushroom body local field potential (LFP) of the locust. What is the contribution of primary afferents in the generation of these dynamics? We addressed this question in two ways. First, we recorded odor-evoked afferent activity in both isolated antennae and intact preparations. Odor-evoked population activity in the antenna and the antennal nerve consisted of a slow potential deflection, similar for many odors. This deflection contained neither oscillatory nor odor-specific slow temporal patterns, whereas simultaneously recorded mushroom body LFPs exhibited clear 20–30 Hz oscillations. This suggests that the temporal patterning of antennal lobe and mushroom body neurons is generated downstream of the olfactory receptor axons. Second, we electrically stimulated arrays of primary afferents in

Numerous studies have suggested that the brain may encode information in the temporal firing pattern of neurons. However, little is known...

In this paper, we extend the Hierarchical Mixtures of Experts (HME) to temporal processing and explore it for a substantial problem, that of text-dependent speaker identification. For a specific multiway classification, we propose a generalized Bernoulli density instead of the multinomial logit density to avoid the instability during training. Time-delay technique is applied for spatio-temporal processing in the HME and a combining scheme is presented for combining multiple time-delay HMEs in order to complete multi-scale analysis for the temporal data. Using the time-delay HME along with the EM algorithm as well as the combination of multiple time-delay HMEs, the speaker identification system has a good performance and yields significantly fast training. We have also addressed some issues about the time-delay techniques in the HME. 1 Introduction Speaker recognition is the process of automatically recognizing who is speaking on the basis of individual information included in speech w...