The recurrent network, containing six continuous-time analog neurons and 42 free parameters (connection strengths and thresholds), is trained to generate time-varying outputs approximating given periodic signals presented to the network. The chip implements a stochastic perturbative algorithm, which observes the error gradient along random directions in the parameter space for error-descent learning. In addition to the integrated learning functions and the generation of pseudo-random perturbations, the chip provides for teacher forcing and long-term storage of the volatile parameters. The network learns a 1 kHz circular trajectory in 100 sec. The chip occupies 2mm x 2mm in a 2JLm CMOS process, and dissipates 1.2 m W. 1 Introduction Exact gradient-descent algorithms for supervised learning in dynamic recurrent networks [1-3] are fairly complex and do not provide for a scalable implementation in a standard 2-D VLSI process. We have implemented a fairly simple and scalable ·Present address: Johns Hopkins University, ECE Dept., Baltimore MD 21218-2686.

What Does the Hippocampus Compute?: A Precis of the 1993 NIPS Workshop Computational models of the hippocampal-region provide an important method for understanding the functional role of this brain system in learning and memory. The presentations in this workshop focused on how modeling can lead to a unified understanding of the interplay among hippocampal physiology, anatomy, and behavior. One approach can be characterized as "top-down" analyses of the neuropsychology of memory, drawing upon brain-lesion studies in animals and humans. Other models take a "bottom-up" approach, seeking to infer emergent computational and functional properties from detailed analyses of circuit connectivity and physiology (see Gluck & Granger, 1993, for a review). Among the issues discussed were: (1) integration of physiological and behavioral theories of hippocampal function, (2) similarities and differences between animal and human studies, (3) representational vs. temporal properties of hippocampaldependent behaviors, (4) rapid vs. incremental learning, (5) mUltiple vs. unitary memory systems, (5) spatial navigation and memory, and (6) hippocampal interaction with other brain systems.

University of Rochester) and ICSIM (lCSI Berkeley) allow the definition of unit types and complex connectivity patterns. On a very high level of abstraction, simulators like tleam (UCSD) allow the easy realization of predefined network architectures (feedforward networks) and leaming algorithms such as backpropagation. Ben Gomes, International Computer Science Institute (Berkeley) introduced the Connectionist Supercomputer 1. The CNSl is a multiprocessor system designed for moderate precision fixed point operations used extensively in connectionist network calculations. Custom VLSI digital processors employ an on-chip vector coprocessor unit tailored for neural network calculations and controlled by RISC scalar CPU.

Abstract: There exist a number of negative results ([J), [BR), [KV]) about learning on neural nets in Valiant's model [V) for probably approximately correct learning ("PAClearning"). These negative results are based on an asymptotic analysis where one lets the number of nodes in the neural net go to infinit.y. Hence this analysis is less adequate for the investigation of learning on a small fixed neural net.

This fact provides a theoretical justification of the widely observed excellent robustness properties of the LMS and backpropagation algorithms. We further discuss some implications of these results.

Recently, Ott, Grebogi and Yorke (OGY) [6] found an effective method to control chaotic systems to unstable fixed points by using only small control forces; however, OGY's method is based on and limited to a linear theory and requires considerable knowledge of the dynamics of the system to be controlled. In this paper we use two radial basis function networks: one as a model of an unknown plant and the other as the controller. The controller is trained with a recurrent learning algorithm to minimize a novel objective function such that the controller can locate an unstable fixed point and drive the system into the fixed point with no a priori knowledge of the system dynamics. Our results indicate that the neural controller offers many advantages over OGY's technique.

We developed a system for finding address blocks on mail pieces that can process four images per second. Besides locating the address block, our system also determines the writing style, handwritten or machine printed, and moreover, it measures the skew angle of the text lines and cleans noisy images. A layout analysis of all the elements present in the image is performed in order to distinguish drawings and dirt from text and to separate text of advertisement from that of the destination address. A speed of more than four images per second is obtained on a modular hardware platform, containing a board with two of the NET32K neural net chips, a SP ARC2 processor board, and a board with 2 digital signal processors. The system has been tested with more than 100,000 images. Its performance depends on the quality of the images, and lies between 85% correct location in very noisy images to over 98% in cleaner images.

This study explores the extent to which a network that learns the temporal relationships within and between the component features of Western tonal music can account for music theoretic and psychological phenomena such as the tonal hierarchy and rhythmic expectancies. Predicted and generated sequences were recorded as the representation of a 153-note waltz melody was learnt by a predictive, recurrent network. The network learned transitions and relations between and within pitch and timing components: accent and duration values interacted in the development of rhythmic and metric structures and, with training, the network developed chordal expectancies in response to the activation of individual tones. Analysis of the hidden unit representation revealed that musical sequences are represented as transitions between states in hidden unit space.

The task of drug activity prediction is to predict the activity of proposed drug compounds by learning from the observed activity of previously-synthesized drug compounds. Accurate drug activity prediction can save substantial time and money by focusing the efforts of chemists and biologists on the synthesis and testing of compounds whose predicted activity is high. If the requirements for highly active binding can be displayed in three dimensions, chemists can work from such displays to design new compounds having high predicted activity. Drug molecules usually act by binding to localized sites on large receptor molecules or large enyzme molecules. One reasonable way to represent drug molecules is to capture the location of their surface in the (fixed) frame of reference of the (hypothesized) binding site.

While exploring to find better solutions, an agent performing online reinforcement learning (RL) can perform worse than is acceptable. In some cases, exploration might have unsafe, or even catastrophic, results, often modeled in terms of reaching'failure' states of the agent's environment. This paper presents a method that uses domain knowledge to reduce the number of failures during exploration. This method formulates the set of actions from which the RL agent composes a control policy to ensure that exploration is conducted in a policy space that excludes most of the unacceptable policies. The resulting action set has a more abstract relationship to the task being solved than is common in many applications of RL. Although the cost of this added safety is that learning may result in a suboptimal solution, we argue that this is an appropriate tradeoff in many problems. We illustrate this method in the domain of motion planning. "'This work was done while the first author was finishing his Ph.D in computer science at the University of Massachusetts, Amherst.