Spotting tasks require detection of target patterns from a background of richly varied non-target inputs. The performance measure of interest for these tasks, called the figure of merit (FOM), is the detection rate for target patterns when the false alarm rate is in an acceptable range. A new approach to training spotters is presented which computes the FOM gradient for each input pattern and then directly maximizes the FOM using b ackpropagati on. This eliminates the need for thresholds during training. It also uses network resources to model Bayesian a posteriori probability functions accurately only for patterns which have a significant effect on the detection accuracy over the false alarm rate of interest. FOM training increased detection accuracy by 5 percentage points for a hybrid radial basis function (RBF) - hidden Markov model (HMM) wordspotter on the credit-card speech corpus.

We study tltt' problem of when to stop If'arning a class of feedforward networks - networks with linear outputs I1PUrOIl and fixed input weights - when they are trained with a gradient descent algorithm on a finite number of examples. Under general regularity conditions, it is shown that there a.re in general three distinct phases in the generalization performance in the learning process, and in particular, the network has hetter gt'neralization pPTformance when learning is stopped at a certain time before til(' global miniIl111lu of the empirical error is reachert. A notion of effective size of a machine is rtefil1e i and used to explain the tradeoff betwf'en the complexity of the marhine and the training error ill the learning process. The study leads nat.urally to a network size selection critt'rion, which turns Ol1t to be a generalization of Akaike's Information Criterioll for the It'arning process. It if; shown that stopping Iparning before tiJt' global minimum of the empirical error has the effect of network size splectioll. 1 INTRODUCTION The primary goal of learning in neural nets is to find a network that gives valid generalization. In achieving this goal, a central issue is the tradeoff between the training error and network complexity. This usually reduces to a problem of network size selection, which has drawn much research effort in recent years. Various principles, theories, and intuitions, including Occam's razor, statistical model selection criteria such as Akaike's Information Criterion (AIC) [11 and many others [5, 1, 10,3,111 all quantitatively support the following PAC prescription: between two machines which have the same empirical error, the machine with smaller VC-dimf'nsion generalizes better. However, it is noted that these methods or criteria do not npcpssarily If'ad to optimal (or llearly optimal) generalization performance.

Catastrophic interference in connectionist networks: Can it be predicted, can it be prevented? Catastrophic forgetting occurs when connectionist networks learn new information, and by so doing, forget all previously learned information. This workshop focused primarily on the causes of catastrophic interference, the techniques that have been developed to reduce it, the effect of these techniques on the networks' ability to generalize, and the degree to which prediction of catastrophic forgetting is possible. The speakers were Robert French, Phil Hetherington (Psychology Department, McGill University, het@blaise.psych.mcgill.ca), French indicated that catastrophic forgetting is at its worst when high representation overlap at the hidden layer combines with significant teacher-output error.

We describe the use of smoothing spline analysis of variance (SS ANOVA) in the penalized log likelihood context, for learning (estimating) the probability p of a '1' outcome, given a training set with attribute vectors and outcomes.

In this paper, we discuss the nature of the time dependence currently employed in our systems using recurrent networks (RNs) and feed-forward multi-layer perceptrons (MLPs). In particular, we introduce local recurrences into a MLP to produce an enhanced input representation. This is in the form of an adaptive gamma filter and incorporates an automatic approach for learning temporal dependencies. We have experimented on a speakerindependent phone recognition task using the TIMIT database. Results using the gamma filtered input representation have shown improvement over the baseline MLP system. Improvements have also been obtained through merging the baseline and gamma filter models.

We show how randomly scrambling the output classes of various fractions of the training data may be used to improve predictive accuracy of a classification algorithm. We present a method for calculating the "noise sensitivity signature" of a learning algorithm which is based on scrambling the output classes. This signature can be used to indicate a good match between the complexity of the classifier and the complexity of the data. Use of noise sensitivity signatures is distinctly different from other schemes to avoid overtraining, such as cross-validation, which uses only part of the training data, or various penalty functions, which are not data-adaptive. Noise sensitivity signature methods use all of the training data and are manifestly data-adaptive and nonparametric. They are well suited for situations with limited training data. 1 INTRODUCTION A major problem of pattern recognition and classification algorithms that learn from a training set of examples is to select the complexity of the model to be trained. How is it possible to avoid an overparameterized algorithm from "memorizing" the training data?

We present a class of feedback control functions which accelerate convergence rates of autonomous nonlinear dynamical systems such as neural network models, without affecting the basic convergence properties (e.g.

In this paper, we show a reciprocal implementation of the "softmax" nonlinearity that is usually used to enforce local competition between neurons [Peterson, 1989]. We show that the circuit is passive and incrementally passive, and we explicitly compute its content and co-content functions. This circuit adds a new element to the library of the analog circuit designer that can be combined with reciprocal constraint boxes [Harris, 1988] and nonlinear resistive fuses [Harris, 1989] to form fast, analog VLSI optimization networks.

This paper describes the use of a convolutional neural network to perform address block location on machine-printed mail pieces. Locating the address block is a difficult object recognition problem because there is often a large amount of extraneous printing on a mail piece and because address blocks vary dramatically in size and shape. We used a convolutional locator network with four outputs, each trained to find a different corner of the address block. A simple set of rules was used to generate ABL candidates from the network output. The system performs very well: when allowed five guesses, the network will tightly bound the address delivery information in 98.2% of the cases.

Several recurrent networks have been proposed as representations for the task of formal language learning. After training a recurrent network recognize a formal 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.