Abstract. This paper focuses on using beam search, a heuristic search algorithm, for pruning state spaces while generating. The original beam search is adapted to the state space generation setting and two new search variants are devised. The resulting framework encompasses some known algorithms, such as A ∗. We also report on two case studies based on an implementation of beam search in μCRL. 1

Further advances in speech recognition heavily depend on the design of architectures which are flexible enough to accomodate very different requirements for the flow of data and hypotheses. These requirements result from the peculiarities of the speech understanding process, which is a complex decision procedure covering a number of different levels of language description. The paper identifies some reasons for assuming additional communicative needs within the recognition system and points out two areas of ongoing research: ffl the design of prototype speech processing modules, which basically rely on inter-module interaction and ffl the development of software-tools for the creation and modification of different architectural layouts. 1 Introduction Speech recognition has made considerable progress during the last decade. Recognition rates have increased with almost a constant ratio. Systems became to a certain degree speaker independent ones and nowadays even accept connected spee...

Search for the global minimum in a molecular energy landscape populated with numerous local minima is a difficult task. Search techniques relevant to such complex spaces can be classified as either global or local. Global search explores the entire space, guaranteeing the global extremum will be found. To accomplish this, the number of samples required grows exponentially with the number of dimensions. Since this is clearly not computationally tractable, global search is impractical in highdimensional spaces. Local search, on the other hand, employs gradient descent to avoid searching the entire exponential space. Gradient descent methods are susceptible to getting stalled in local minima and consequently, no guarantees can be made about finding the global minimum. We propose a middle ground that minimizes the effects of exponential space and local minima by integrating domain knowledge and information generated during search into a model, and then using this model to focus computation on regions of increasing relevance. Directing resources to multiple relevant regions prevents oversampling local minima. At the same time the exploration of only significant regions avoids the intractable computational requirements of high-dimensional spaces. The proposed method, called Model-Based Search (MBS), is compared to the local search method Monte Carlo as implemented in Rosetta- currently considered the best computational protein structure prediction method. The results indicate that MBS is significantly better at finding lower energy minima than the Monte Carlo technique implemented as part of Rosetta. This effect is amplified as the dimensionality of the search space increases. 1

Abstract not found

Beam search algorithms for the early/tardy scheduling problem with release dates

This paper presents a design methodology for optimizing Viterbi beam search in HMM (hidden Markov model) decoding for isolated-word speech recognition. The proposed data-driven method can effectively identify a near-optimal beam-search ranking curve (BSRC) that can reduce the computation time to an acceptable amount while minimizing the reduction in recognition rate based on a set of sample data. Experimental results based on the most famous 300 poems in Tang Dynasty of China demonstrate the feasibility of the proposed approach. 1.

The stack decoder is an attractive algorithm for con-trolling the acoustic and language model matching in a continuous speech recognizer. It implements a best-first tree search of the language to find the best match to both the language model and the observed speech. This paper describes a method for performing the optimal A* search which guarantees to find the most likely path (rec-ognized sentence) while extending the minimum number of stack entries. A tree search, however, is exponential in the number of words. A second algorithm is presented which linearizes the search at the cost of approximating some of the path likelihoods.

In this paper we present the ISADORA system which provides highly flexible speech recognition based on HMM technology together with an hierarchical representation of speech units. Markov model topologies, subword unit inventories, regular grammars expressed in finite-state or phrase structure style, and even the analysis tasks themselves are explicitly represented by the nodes of a large speech unit network. Thus, nothing that can be "said in the language of Markov models" needs to be hard-wired in the program code. In contrast to traditional compiled network recognizers, units, grammars, and tasks may be created or modified at analysis time, and the outcome of the decoding process is a structured symbolic description of the sensory input. Our architecture has proven extremely useful in prototyping new kinds of subword units. Besides generalized triphones and context-freezing units, a new subword speech unit for automatic speech recognition has been implemented. The so-called polyphone...

In this paper we describe the BBN Byblos 10x real time system used for the 1998 Hub-4 English tests. Given our state of the art primary system [1] running at 230 times real time (230 xRT) we show that eliminating and approximating many computationally expensive components speeds up the system by a factor of 23 with a relative loss in WER of 18%. This is accomplished without retraining or changing the primary system structure. The components of the primary system that are refined include segmentation, adaptation, decoding, cross-word rescoring with adaptation, and system combination. The time saving algorithms used include fast Gaussian computation, grammar spreading, nbest tree rescoring, and block diagonal adaptation. 1. INTRODUCTION Large vocabulary continuous speech recognition requires a considerable amount of computation. The amount of computation depends to a large degree on the quality of speech, with the computation increasing by a significant factor for more natural speech. ...

This .paper describes the Intelligent Management System (IMS) project, which is part of the Factory of the Future project in the Robotics Institute of Carnegie-Mellon University.. IMS is a long term project concerned with applying artificial intelligence techniques in aiding professionals and managers in their day to day tasks. This report discusses both the long term goals of IMS, and current research. It describes research in the medeling of organizations, constraint- based job-shop scheduling, organization simulation, user interfaces, and system architecture. Examples of working systems are provided.