, wire-free accelerometers on 20 activities using datasets annotated by the subjects themselves.

Distributed computing systems are being built and used more and more frequently. This distributod computing revolution makes the reliability of distributed systems an important concern. It is fairly well-understood how to connect hardware so that most components can continue to work when others are broken, and thus increase the reliability of a system as a whole. This report addressos the issue of providing software for reliable distributed systems. In particular, we examine how to program a system so that the software continues to work in tho face of a variety of failures of parts of the system. The design presented

describe a method for statistical modeling based on maximum entropy. We present a maximum-likelihood approach for automatically constructing maximum entropy models and describe how to implement this approach efficiently, using as examples several problems in natural language processing.

conclude by presenting some of the benefits of our approach to software architecture, sum-marizing our contributions, and relating our approach to other current work.

of Scientific Research, Air Force Materiel Command, USAF, under grant number F49620-98-1-0061; and by the National Science Foundation under Grant Number CCR-9701973. The U.S. Government is authorized to reproduce and distribute reprints for Governmental purposes notwithstanding any copyright annotation thereon

This article presents a measure of semantic similarityinanis-a taxonomy based on the notion of shared information content. Experimental evaluation against a benchmark set of human similarity judgments demonstrates that the measure performs better than the traditional edge-counting approach

restrictive assumptions can perform even better. In this paper we evaluate approaches for inducing classifiers from data, based on the theory of learning Bayesian networks. These networks are factored representations of probability distributions that generalize the naive Bayesian classifier and explicitly

software we will gain access to more data, but not improve access to the type and quality of information needed for decision making. To develop the concepts needed for future information systems we model information processing as an interaction of data and knowledge. This model provides criteria for a high

and data dependences. Program transformations that require interaction of the two dependence types can also be easily handled with our representation. As an example, an incremental approach to modifying data dependences resulting from branch deletion or loop unrolling is intro-duced. The PDG supports

all be understood in terms of exact or approximate forms of these variational representations. The variational approach provides a complementary alternative to Markov chain Monte Carlo as a general source of approximation methods for inference in large-scale statistical models.