A Hidden-Articulator Markov Model (HAMM) is a Hidden Markov Model (HMM) in which each state represents an articulatory configuration. Articulatory knowledge, known to be useful for speech recognition [4], is represented by specifying a mapping of phonemes to articulatory configurations; vocal tract dynamics are represented via transitions between articulatory configurations. In previous work [13], we extended the articulatory-feature model introduced by Erler [7] by using diphone units and a new technique for model initialization. By comparing it with a purely random model, we showed that the HAMM can take advantage of articulatory knowledge. In this paper, we extend that work in three ways. First, we decrease the number of parameters, making it comparable in size to standard HMMs. Second, we evaluate our model in noisy contexts, verifying that articulatory knowledge can provide benefits in adverse acoustic conditions. Third, we use a corpus of sideby -side speech and articulator tra...

e-mail secre-

Symbolic representations have been used successfully in off-line planning algorithms for Markov decision processes. We show that they can also improve the performance of online planners. In addition to reducing computation time, symbolic generalization can reduce the amount of costly real-world interactions required for convergence. We introduce Symbolic Real-Time Dynamic Programming (or sRTDP), an extension of RTDP. After each step of on-line interaction with an environment, sRTDP uses symbolic modelchecking techniques to generalizes its experience by updating a group of states rather than a single state. We examine two heuristic approaches to dynamic grouping of states and show that they accelerate the planning process significantly in terms of both CPU time and the number of steps of interaction with the environment.

Automatic continuous speech recognition (CSR) is sufficiently

Hidden Markov models assume that observations in time series data stem from some hidden process that can be compactly represented as a Markov chain. We generalize this model by assuming that the observed data stems from multiple hidden processes, whose outputs interleave to form the sequence of observations. Exact inference in this model is NP-hard. However, a tractable and effective inference algorithm is obtained by extending structured approximate inference methods used in factorial hidden Markov models. The proposed model is evaluated in an activity recognition domain, where multiple activities interleave and together generate a stream of sensor observations. It is shown to be more accurate than a standard hidden Markov model in this domain.

This study investigated recognition systems that combine loosely coupled modalities, integrating eye movements in an Automatic Speech Recognition (ASR) system as an exemplar. A probabilistic framework for combining modalities was formalised and applied to the specific case of integrating eye movement and speech. A corpus of a matched eye movement and related spontaneous conversational British English speech for a visual-based, goal-driven task was collected. This corpus enabled the relationship between the modalities to be verified. Robust extraction of visual attention from eye movement data was investigated using Hidden Markov Models and Hidden Semi-Markov Models. Gaze-contingent ASR systems were developed from a research-grade baseline ASR system by redistributing language model probability mass according to the visual attention. The best performing systems maintained the Word Error Rates but showed an increase in the Figure of Merit- a measure of the keyword spotting accuracy and integration success. The core values of this work may be useful for developing robust multimodal decoding system functions.

The modeling of high level semantic events from low level sensor signals is important in order to understand distributed phenomena. For such content-modeling purposes, transformation of numeric data into symbols and the modeling of resulting symbolic sequences can be achieved using statistical models—Markov Chains (MCs) and Hidden Markov Models (HMMs). We consider the problem of distributed indexing and semantic querying over such sensor models. Specifically, we are interested in efficiently answering (i) range queries: return all sensors that have observed an unusual sequence of symbols with a high likelihood, (ii) top-1 queries: return the sensor that has the maximum probability of observing a given sequence, and (iii) 1-NN queries: return the sensor (model) which is most similar to a query model. All the above queries can be answered at the centralized base station, if each sensor transmits its model to the base station. However, this is communicationintensive. We present a much more efficient alternative—a distributed index structure, MIST (Model-based Index STructure), and accompanying algorithms for answering the above queries. MIST aggregates two or more constituent models into a single composite model, and constructs an in-network hierarchy over such composite models. We develop two kinds of composite models: the first kind captures the average behavior of the underlying models and the second kind captures the extreme behaviors of the underlying models. Using the index parameters maintained at the root of a subtree, we bound the probability of observation of a query sequence from a sensor in the subtree. We also bound the distance of a query model to a sensor model using these parameters. Extensive experimental evaluation on both real-world and synthetic data sets show that the MIST schemes scale well in terms of network size and number of model states. We also show its superior performance over the centralized schemes in terms of update, query, and total communication costs.

Robust sequence prediction is an essential component of an intelligent agent acting in a dynamic world. We consider the case of near-future event prediction by an online learning agent operating in a non-stationary environment. The challenge for a learning agent under these conditions is to exploit the relevant experience from a limited environmental event history while preserving flexibility. We propose a novel time/space efficient method for learning temporal sequences and making short-term predictions. Our method operates on-line, requires few exemplars, and adapts easily and quickly to changes in the underlying stochastic world model. Using a short-term memory of recent observations, the method maintains a dynamic space of candidate hypotheses in which the growth of the space is systematically and dynamically pruned using an entropy measure over the observed predictive quality of each candidate hypothesis. The method compares well against Markov-chain predictions, and adapts faster than learned Markov-chain models to changes in the underlying distribution of events. We demonstrate the method using both synthetic data and empirical experience from a gameplaying scenario with human opponents.

We investigate a general approach to approximately learning a compact and structured representation of the transition model for Factorial Markov Decision Processes (FMDPs). FMDPs are based on mixed memory Markov models, in which the transition probabilities are factored into a mixture of terms depending on each state variable.

In this paper we present Dynamic Bayesian Networks (DBN) as computational framework for the analysis of dynamic systems. We first give a short overview of the work on state estimation and system identification. Then we present two application where the DBN is used: robot localization and the tracking of multiple persons with multiple camera’s. 1