This is a summary of results with Dyna, a class of architectures for intelligent systems based on approximating dynamic programming methods. Dyna architectures integrate trial-and-error (reinforcement) learning and execution-time planning into a single process operating alternately on the world and on a learned forward model of the world. We describe and show results for two Dyna architectures, Dyna-AHC and Dyna-Q. Using a navigation task, results are shown for a simple Dyna-AHC system which simultaneously learns by trial and error, learns a world model, and plans optimal routes using the evolving world model. We show that Dyna-Q architectures (based on Watkins's Q-Iearning) are easy to adapt for use in changing environments.

This paper presents a neural network (NN) approach to the problem of stereopsis. The correspondence problem (finding the correct matches between the pixels of the epipolar lines of the stereo pair from amongst all the possible matches) is posed as a non-iterative many-to-one mapping. A two-layer feed forward NN architecture is developed to learn and code this nonlinear and complex mapping using the back-propagation learning rule and a training set. The important aspect of this technique is that none of the typical constraints such as uniqueness and continuity are explicitly imposed. All the applicable constraints are learned and internally coded by the NN enabling it to be more flexible and more accurate than the existing methods. The approach is successfully tested on several randomdot stereograms. It is shown that the net can generalize its learned mapping to cases outside its training set. Advantages over the Marr-Poggio Algorithm are discussed and it is shown that the NN performance is superIOr.

This work presents an Attractor Neural Network (ANN) model of Recall and Recognition. It is shown that an ANN model can qualitatively account for a wide range of experimental psychological data pertaining to the these two main aspects of memory access. Certain psychological phenomena are accounted for, including the effects of list-length, wordfrequency, presentation time, context shift, and aging. Thereafter, the probabilities of successful Recall and Recognition are estimated, in order to possibly enable further quantitative examination of the model. 1 Motivation The goal of this paper is to demonstrate that a Hopfield-based [Hop82] ANN model can qualitatively account for a wide range of experimental psychological data pertaining to the two main aspects of memory access, Recall and Recognition. Recall is defined as the ability to retrieve an item from a list of items (words) originally presented during a previous learning phase, given an appropriate cue (cued RecalQ, or spontaneously (free RecalQ. Recognition is defined as the ability to successfully acknowledge that a certain item has or has not appeared in the tutorial list learned before. The main prospects of ANN modeling is that some parameter values, that in former, 'classical' models of memory retrieval (see e.g.

Simple classical spin models well-known to physicists as the ANNNI and Heisenberg XY Models. in which long-range interactions occur in a pattern given by the Mexican Hat operator.

In response to a puff of wind, the American cockroach turns away and runs. The circuit underlying the initial turn of this escape response consists of three populations of individually identifiable nerve cells and appears to employ distributed representations in its operation. We have reconstructed several neuronal and behavioral properties of this system using simplified neural network models and the backpropagation learning algorithm constrained by known structural characteristics of the circuitry. In order to test and refine the model, we have also compared the model's responses to various lesions with the insect's responses to similar lesions.

We describe a CMOS neural net chip with a reconfigurable network architecture. It contains 32,768 binary, programmable connections arranged in 256 'building block' neurons. Several'building blocks' can be connected to form long neurons with up to 1024 binary connections or to form neurons with analog connections. Single-or multi-layer networks can be implemented with this chip. We have integrated this chip into a board system together with a digital signal processor and fast memory.

Richard S. Zemel and Geoffrey E. Hinton Department of Computer Science University of Toronto Toronto, ONT M5S lA4 Abstract Using an unsupervised learning procedure, a network is trained on an ensemble of images of the same two-dimensional object at different positions, orientations and sizes. Each half of the network "sees" one fragment of the object, and tries to produce as output a set of 4 parameters that have high mutual information with the 4 parameters output by the other half of the network. Given the ensemble of training patterns, the 4 parameters on which the two halves of the network can agree are the position, orientation, and size of the whole object, or some recoding of them. After training, the network can reject instances of other shapes by using the fact that the predictions made by its two halves disagree. If two competing networks are trained on an unlabelled mixture of images of two objects, they cluster the training cases on the basis of the objects' shapes, independently of the position, orientation, and size. 1 INTRODUCTION A difficult problem for neural networks is to recognize objects independently of their position, orientation, or size.

Although color TV is an established technology, there are a number of longstanding problems for which neural networks may be suited. Impulse noise is such a problem, and a modular neural network approach is presented in this paper. The training and analysis was done on conventional computers, while real-time simulations were performed on a massively parallel computer called the Princeton Engine. The network approach was compared to a conventional alternative, a median filter. Real-time simulations and quantitative analysis demonstrated the technical superiority of the neural system. Ongoing work is investigating the complexity and cost of implementing this system in hardware.

John S. Denker and Yann leCun AT&T Bell Laboratories Holmdel, NJ 07733 Abstract (1) The outputs of a typical multi-output classification network do not satisfy the axioms of probability; probabilities should be positive and sum to one. This problem can be solved by treating the trained network as a preprocessor that produces a feature vector that can be further processed, for instance by classical statistical estimation techniques. It is particularly useful to combine these two ideas: we implement the ideas of section 1 using Parzen windows, where the shape and relative size of each window is computed using the ideas of section 2. This allows us to make contact between important theoretical ideas (e.g. the ensemble formalism) and practical techniques (e.g. Our results also shed new light on and generalize the well-known "soft max" scheme. For example, in speech recognition, these numbers represent the probability of C different phonemes; the probabilities of successive segments can be combined using a Hidden Markov Model.

A new class of data structures called "bumptrees" is described. These structures are useful for efficiently implementing a number of neural network related operations. An empirical comparison with radial basis functions is presented on a robot ann mapping learning task.