Tree-structured probabilistic models admit simple, fast inference. However, they are not well suited to phenomena such as occlusion, where multiple components of an object may disappear simultaneously. Mixtures of trees appear to address this problem, at the cost of representing a large mixture. We demonstrate an efficient and compact representation of this mixture, which admits simple learning and inference algorithms. We use this method to build an automated tracker for Muybridge sequences of a variety of human activities. Tracking is difficult, because the temporal dependencies rule out simple inference methods. We show how to use our model for efficient inference, using a method that employs alternate spatial and temporal inference. The result is a tracker that (a) uses a very loose motion model, and so can track many different activities at a variable frame rate and (b) is entirely automatic. 1.

We investigate the problem of energy-efficient broadcast routing over wireless static adhoc network where host mobility is not involved. We define the lifetime of a network as the duration of time until the first node failure due to battery depletion. We provide a globally optimal solution to the problem of maximizing a static network lifetime through a graph theoretic approach. We also provide extensive comparative simulation studies.

The trend towards very large DNA sequencing projects, such as those being undertaken as part of the human genome initiative, necessitates the development of efficient and precise algorithms for assembling a long DNA sequence from the fragments obtained by shotgun sequencing or other methods. The sequence reconstruction problem that we take as our formulation of DNA sequence assembly is a variation of the shortest common superstring problem, complicated by the presence of sequencing errors and reverse complements of fragments. Since the simpler superstring problem is NP-hard, any efficient reconstruction procedure must resort to heuristics. In this paper, however, a four phase approach based on rigorous design criteria is presented, and has been found to be very accurate in practice. Our method is robust in the sense that it can accommodate high sequencing error rates and list a series of alternate solutions in the event that several appear equally good. Moreover it uses a limited form ...

A face detection system is presented. A new classi cation method using foreststructured Bayesian networks is used. The method is used in an aggregated classi er to discriminate face from non-face patterns. The process of generating non-face patterns is integrated with the construction of the aggregated classi er. The face detection system performs well in comparison with other well-known methods.

Comparative genome hybridization (CGH) is a laboratory method to measure gains and losses of chromosomal regions in tumor cells. It is believed that DNA gains and losses in tumor cells do not occur entirely at random, but partly through some flow of causality. Models that relate tumor progression to the occurrence of DNA gains and losses could be very useful in hunting cancer genes and in cancer diagnosis. We lay some mathematical foundations for inferring a model of tumor progression from a CGH data set. We consider a class of tree models that are more general than a path model that has been developed for colorectal cancer. We derive a tree model inference algorithm based on the idea of a maximum-weight branching in a graph, and we show that under plausible assumptions our algorithm infers the correct tree. We have implemented our methods in software, and we illustrate with a CGH data set for renal cancer.

This thesis deals with a class of Lagrangian relaxation based algorithms developed in the computer science community in last couple of decades. We present a unified framework for designing such algorithms for a large family of convex programming problems. Our algorithms are based on exponential potential functions and given any 2 (0; 1), compute (1 + )-approximate solutions in number of iterations proportional to .

In this paper, we consider energy-e#cient gathering of correlated data in sensor networks. We focus on single-input coding strategies in order to aggregate correlated data. For foreign coding we propose the MEGA algorithm which yields a minimum-energy data gathering topology in O time. We also consider self-coding for which the problem of finding an optimal data gathering tree was recently shown to be NP-complete; with LEGA, we present the first approximation algorithm for this problem with approximation ratio 2(1 + # 2) and running time O(m + n log n).

Delta compression techniques are commonly used to succinctly represent an updated version of a file with respect to an earlier one. In this paper, we study the use of delta compression in a somewhat different scenario, where we wish to compress a large collection of (more or less) related files by performing a sequence of pairwise delta compressions. The problem of finding an optimal delta encoding for a collection of files by taking pairwise deltas can be reduced to the problem of computing a branching of maximum weight in a weighted directed graph, but this solution is inefficient and thus does not scale to larger file collections. This motivates us to propose a framework for cluster-based delta compression that uses text clustering techniques to prune the graph of possible pairwise delta encodings. To demonstrate the efficacy of our approach, we present experimental results on collections of web pages. Our experiments show that cluster-based delta compression of collections provides significant improvements in compression ratio as compared to individually compressing each file or using tar+gzip, at a moderate cost in efficiency.

Efficient detection of objects in images is complicated by variations of object appearance due to intra-class object differences, articulation, lighting, occlusions, and aspect variations. To reduce the search required for detection, we employ the bottom-up approach where we find candidate image features and associate some of them with parts of the object model. We represent objects as collections of local features, and would like to allow any of them to be absent, with only a small subset sufficient for detection; furthermore, our model should allow efficient correspondence search. We propose a model, Mixture of Trees, that achieves these goals. With a mixture of trees, we can model the individual appearances of the features, relationships among them, and the aspect, and handle occlusions. Independences captured in the model make efficient inference possible. In our earlier work, we have shown that mixtures of trees can be used to model objects with a natural tree structure, in the context of human tracking. Now we show that a natural tree structure is not required, and use a mixture of trees for both frontal and view-invariant face detection. We also show that by modeling faces as collections of features we can establish an intrinsic coordinate frame for a face, and estimate the out-of-plane rotation of a face.

Delta compression and remote file synchronization techniques are concerned with efficient file transfer over a slow communication link in the case where the receiving party already has a similar file (or files). This problem arises naturally, e.g., when distributing updated versions of software over a network or synchronizing personal files between different accounts and devices. More generally, the problem is becoming increasingly common in many networkbased applications where files and content are widely replicated, frequently modified, and cut and reassembled in different contexts and packagings.