Michael C. Hughes

JOINT MODELING OF MULTIPLE TIME SERIES VIA THE BETA PROCESS WITH APPLICATION TO MOTION CAPTURE SEGMENTATION

We propose a Bayesian nonparametric approach to the problem of jointly modeling multiple related time series. Our model discovers a latent set of dynamical behaviors shared among the sequences, and segments each time series into regions defined by a subset of these behaviors. Using a beta process prior, the size of the behavior set and the sharing pattern are both inferred from data. We develop Markov chain Monte Carlo (MCMC) methods based on the Indian buffet process representation of the predictive distribution of the beta process. Our MCMC inference algorithm efficiently adds and removes behaviors via novel split-merge moves as well as data-driven birth and death proposals, avoiding the need to consider a truncated model. We demonstrate promising results on unsupervised segmentation of human motion capture data. 1. Introduction. Classical time series analysis has generally focused on the study of a single (potentially multivariate) time series. Instead, we consider analyzing collections of related time series, motivated by the increasing abundance of such data in many domains. In this work we explore this problem by considering time series produced by motion capture sensors on the joints of people performing exercise routines. An individual recording provides a multivariate time series that can be segmented into types of exercises (e.g., jumping jacks, arm-circles, and twists). Each exercise type describes locally coherent and simple dynamics that persist over a segment of time. We have such motion capture recordings from multiple individuals, each of whom performs some subset of a global set of exercises, as shown in Figure1. Our goal is to discover the set of global exercise types (“behaviors”) and their occurrences in each individual’s data stream. We would like to take advantage of the overlap between individuals: if a jumping-jack behavior is discovered in one sequence, then it can be used to model data for other individuals.

Right for the Right Reasons: Training Differentiable Models by Constraining their Explanations

Neural networks are among the most accurate supervised learning methods in use today, but their opacity makes them difficult to trust in critical applications, especially when conditions in training differ from those in test. Recent work on explanations for black-box models has produced tools (e.g. LIME) to show the implicit rules behind predictions, which can help us identify when models are right for the wrong reasons. However, these methods do not scale to explaining entire datasets and cannot correct the problems they reveal. We introduce a method for efficiently explaining and regularizing differentiable models by examining and selectively penalizing their input gradients, which provide a normal to the decision boundary. We apply these penalties both based on expert annotation and in an unsupervised fashion that encourages diverse models with qualitatively different decision boundaries for the same classification problem. On multiple datasets, we show our approach generates faithful explanations and models that generalize much better when conditions differ between training and test.

Nonparametric discovery of activity patterns from video collections

We propose a nonparametric framework based on the beta process for discovering temporal patterns within a heterogenous video collection. Starting from quantized local motion descriptors, we describe the long-range temporal dynamics of each video via transitions between a set of dynamical behaviors. Bayesian nonparametric statistical methods allow the number of such behaviors and the subset exhibited by each video to be learned without supervision. We extend the earlier beta process HMM in two ways: adding data-driven MCMC moves to improve inference on realistic datasets and allowing global sharing of behavior transition parameters. We illustrate discovery of intuitive and useful dynamical structure, at various temporal scales, from videos of simple exercises, recipe preparation, and Olympic sports. Segmentation and retrieval experiments show the benefits of our nonparametric approach.