The aim of the project was to make a signature verification system based on the NCR 5990 Signature Capture Device (a pen-input tablet) and to use 80 bytes or less for signature feature storage in order that the features can be stored on the magnetic strip of a credit-card. Verification using a digitizer such as the 5990, which generates spatial coordinates as a function of time, is known as dynamic verification. Much research has been carried out on signature verification. Function-based methods, which fit a function tothe pen trajectory, have been found to lead to higher performance while parameter-based methods, which extract some number of parameters from a signa-737 738 Bromley, Guyon, Le Cun, Sackinger, and Shah ture, make a lower requirement on memory space for signature storage (see Lorette and Plamondon (1990) for comments). We chose to use the complete time extent of the signature, with the preprocessing described below, as input to a neural network, andto allow the network to compress the information.

In solving these tasks, one is faced with a large variety of learning algorithms and a vast selection of possible network architectures. After all the training, how does one know which is the best network? This decision is further complicated by the fact that standard techniques can be severely limited by problems such as over-fitting, data sparsity and local optima. The usual solution to these problems is a winner-take-all cross-validatory model selection. However, recent experimental and theoretical work indicates that we can improve performance by considering methods for combining neural networks.

While exploring to find better solutions, an agent performing online reinforcementlearning (RL) can perform worse than is acceptable. Insome 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. Thismethod 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 tradeoffin 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.

Olshausen Department of Anatomy and Neurobiology Washington University School of Medicine St. Louis, MO 63110 The purpose of this workshop was to discuss both recent experimental findings and computational models of the neurobiological implementation of selective attention. Recent experimental results were presented in two of the four presentations given (C.E. Connor, Washington University and B.C. Motter, SUNY and V.A. Medical Center, Syracuse), while the other two talks were devoted to computational models (E. Connor presented the results of an experiment in which the receptive field profiles of V 4 neurons were mapped during different states of attention in an awake, behaving monkey. The attentional focus was manipulated in this experiment by altering the position of a behaviorally relevant ring-shaped stimulus.

Several recurrent networks have been proposed as representations for the task of formal language learning. After training a recurrent network recognize aformal language or predict the next symbol of a sequence, the next logical step is to understand the information processing carried out by the network. Some researchers have begun to extracting finite state machines from the internal state trajectories of their recurrent networks. This paper describes how sensitivity to initial conditions and discrete measurements can trick these extraction methods to return illusory finite state descriptions.

Survival is enhanced by an ability to predict the availability of food, the likelihood of predators, and the presence of mates. We present a concrete model that uses diffuse neurotransmitter systems to implement a predictive version of a Hebb learning rule embedded in a neural architecture basedon anatomical and physiological studies on bees. The model captured the strategies seen in the behavior of bees and a number of other animals when foraging in an uncertain environment. The predictive model suggests a unified way in which neuromodulatory influences can be used to bias actions and control synaptic plasticity. Successful predictions enhance adaptive behavior by allowing organisms to prepare for future actions,rewards, or punishments. Moreover, it is possible to improve upon behavioral choices if the consequences of executing different actions can be reliably predicted. Although classicaland instrumental conditioning results from the psychological literature [1] demonstrate that the vertebrate brain is capable of reliable prediction, how these predictions are computed in brains is not yet known. The brains of vertebrates and invertebrates possess small nuclei which project axons throughout large expanses of target tissue and deliver various neurotransmitters such as dopamine, norepinephrine, and acetylcholine [4]. The activity in these systems may report on reinforcing stimuli in the world or may reflect an expectation of future reward [5, 6,7,8].

The Minimum Description Length principle (MDL) can be used to train the hidden units of a neural network to extract a representation thatis cheap to describe but nonetheless allows the input to be reconstructed accurately. We show how MDL can be used to develop highly redundant population codes. Each hidden unit has a location in a low-dimensional implicit space. If the hidden unit activities form a bump of a standard shape in this space, they can be cheaply encoded by the center ofthis bump. So the weights from the input units to the hidden units in an autoencoder are trained to make the activities form a standard bump.

This inverse dynamics model allows the use of a faster speech mot.or control scheme, which can be applied to phoneme-tospeech synthesisvia musclo-skeletal system dynamics, or to future use in speech recognition. The forward acoustic model, which is the mapping from articulator trajectories t.o the acoustic parameters, was improved by adding velocity and voicing information inputs to distinguish acollst.ic

Critical features of a domain (such as invariance under translation, rotation, and permu- Clustering with a Domain-Specific Distance Measure 103 tation) are captured within the clustering procedure, rather than reflected in the properties of feature sets created prior to clustering. The distance measure and learning problem are formally described as nested objective functions. We derive an efficient algorithm by using optimization techniques that allow us to divide up the objective function into parts which may be minimized in distinct phases. The algorithm has accurately recreated 10 prototypes from a randomly generated sample database of 100 images consisting of 20 points each in 120 experiments. Finally, by incorporating permutation invariance in our distance measure, we have a technique that we may be able to apply to the clustering of graphs. Our goal is to develop measures which will enable the learning of objects with shape or structure. Acknowledgements This work has been supported by AFOSR grant F49620-92-J-0465 and ONR/DARPA grant N00014-92-J-4048.

We show how an "Elman" network architecture, constructed from recurrently connected oscillatory associative memory network modules, canemploy selective "attentional" control of synchronization to direct the flow of communication and computation within the architecture to solve a grammatical inference problem. Previously we have shown how the discrete time "Elman" network algorithm can be implemented in a network completely described by continuous ordinary differential equations. The time steps (machine cycles)of the system are implemented by rhythmic variation (clocking) of a bifurcation parameter. In this architecture, oscillation amplitudecodes the information content or activity of a module (unit), whereas phase and frequency are used to "softwire" the network. Only synchronized modules communicate by exchanging amplitudeinformation; the activity of non-resonating modules contributes incoherent crosstalk noise. Attentional control is modeled as a special subset of the hidden modules with ouputs which affect the resonant frequencies of other hidden modules. They control synchrony among the other modules anddirect the flow of computation (attention) to effect transitions betweentwo subgraphs of a thirteen state automaton which the system emulates to generate a Reber grammar. The internal crosstalk noise is used to drive the required random transitions of the automaton.