In this paper, we propose a new unsupervised approach for word segmentation. The core idea of our approach is a novel word induction criterion called WordRank, which estimates the goodness of word hypotheses (character or phoneme sequences). We devise a method to derive exterior word boundary information from the link structures of adjacent word hypotheses and incorporate interior word boundary information to complete the model. In light of WordRank, word segmentation can be modeled as an optimization problem. A Viterbi-styled algorithm is developed for the search of the optimal segmentation. Extensive experiments conducted on phonetic transcripts as well as standard Chinese and Japanese data sets demonstrate the effectiveness of our approach. On the standard Brent version of Bernstein-Ratner corpora, our approach outperforms the state-of-the-art Bayesian models by more than 3%. Plus, our approach is simpler and more efficient than the Bayesian methods. Consequently, our approach is more suitable for real-world applications.

Computer Go presents a challenging problem for machine learning agents. With the number of possible board states estimated to be larger than the number of hydrogen atoms in the universe, learning effective policies or board evaluation functions is extremely difficult. In this paper we describe Cortigo, a system that efficiently and autonomously learns useful generalizations for large state-space classification problems such as Go. Cortigo uses a hierarchical generative model loosely related to the human visual cortex to recognize Go board positions well enough to suggest promising next moves. We begin by briefly describing and providing motivation for research in the computer Go domain. We describe Cortigo’s ability to learn predictive models based on large subsets of the Go board and demonstrate how using Cortigo’s learned models as additive knowledge in a state-of-the-art computer Go player (Fuego) significantly improves its playing strength.

We propose a novel probabilistic model for collaborative filtering (CF), called SRMCoFi, which seamlessly integrates both linear and bilinear random effects into a principled framework. The formulation of SRMCoFi is supported by both social psychological experiments and statistical theories. Not only can many existing CF methods be seen as special cases of SRMCoFi, but it also integrates their advantages while simultaneously overcoming their disadvantages. The solid theoretical foundation of SRMCoFi is further supported by promising empirical results obtained in extensive experiments using real CF data sets on movie ratings.

This paper studies how automated agents can persuade humans to behave in certain ways. The motivation behind such agent's behavior resides in the utility function that the agent's designer wants to maximize and which may be different from the user's utility function. Specifically, in the strategic settings studied, the agent provides correct yet partial information about a state of the world that is unknown to the user but relevant to his decision. Persuasion games were designed to study interactions between automated players where one player sends state information to the other to persuade it to behave in a certain way. We show that this game theory based model is not sufficient to model human-agent interactions, since people tend to deviate from the rational choice. We use machine learning to model such deviation in people from this game theory based model. The agent generates a probabilistic description of the world state that maximizes its benefit and presents it to the users. The proposed model was evaluated in an extensive empirical study involving road selection tasks that differ in length, costs and congestion. Results showed that people's behavior indeed deviated significantly from the behavior predicted by the game theory based model. Moreover, the agent developed in our model performed better than an agent that followed the behavior dictated by the game-theoretical models.

Dependency networks are a compelling alternative to Bayesian networks for learning joint probability distributions from data and using them to compute probabilities. A dependency network consists of a set of conditional probability distributions, each representing the probability of a single variable given its Markov blanket. Running Gibbs sampling with these conditional distributions produces a joint distribution that can be used to answer queries, but suffers from the traditional slowness of sampling-based inference. In this paper, we observe that the mean field update equation can be applied to dependency networks, even though the conditional probability distributions may be inconsistent with each other. In experiments with learning and inference on 12 datasets, we demonstrate that mean field inference in dependency networks offers similar accuracy to Gibbs sampling but with orders of magnitude improvements in speed. Compared to Bayesian networks learned on the same data, dependency networks offer higher accuracy at greater amounts of evidence. Furthermore, mean field inference is consistently more accurate in dependency networks than in Bayesian networks learned on the same data.

We describe the Fourier basis, a linear value function approximation scheme based on the Fourier series. We empirically demonstrate that it performs well compared to radial basis functions and the polynomial basis, the two most popular fixed bases for linear value function approximation, and is competitive with learned proto-value functions.

We consider a learning setting of importance to large scale machine learning: potentially unlimited data arrives sequentially, but only a small fraction of it is labeled. The learner cannot store the data; it should learn from both labeled and unlabeled data, and it may also request labels for some of the unlabeled items. This setting is frequently encountered in real-world applications and has the characteristics of online, semi-supervised, and active learning. Yet previous learning models fail to consider these characteristics jointly. We present OASIS, a Bayesian model for this learning setting. The main contributions of the model include the novel integration of a semi-supervised likelihood function, a sequential Monte Carlo scheme for efficient online Bayesian updating, and a posterior-reduction criterion for active learning. Encouraging results on both synthetic and real-world optical character recognition data demonstrate the synergy of these characteristics in OASIS.

Ensemble classification methods that independently construct component models (e.g., bagging) improve accuracy over single models by reducing the error due to variance. Some work has been done to extend ensemble techniques for classification in relational domains by taking relational data characteristics or multiple link types into account during model construction. However, since these approaches follow the conventional approach to ensemble learning, they improve performance by reducing the error due to variance in learning. We note however, that variance in inference can be an additional source of error in relational methods that use collective classification, since inferred values are propagated during inference. We propose a novel ensemble mechanism for collective classification that reduces  both learning and inference variance, by incorporating prediction averaging into the collective inference process itself. We show that our proposed method significantly outperforms a straightforward relational ensemble baseline on both synthetic and real-world datasets.

Categories are often organized into hierarchical taxonomies, that is, tree structures where each node represents a labeled category, and a node's parent and children are, respectively, the category's supertype and subtypes. A natural question is whether it is possible to reconstruct category taxonomies in cases where we are not given explicit information about how categories are related to each other, but only a sample of observations of the members of each category. In this paper, we introduce a nonparametric Bayesian model of multi-level category learning, an extension of the hierarchical Dirichlet process (HDP) that we call the tree-HDP. We demonstrate the ability of the tree-HDP to reconstruct simulated datasets of artificial taxonomies, and show that it produces similar performance to human learners on a taxonomy inference task.

Peer reviewing is the key ingredient of evaluating the quality of scientific work. Based on the review scores assigned by the individual reviewers to the submissions, program committees of conferences and journal editors decide which papers to accept for publication and which to reject. However, some reviewers may be more rigorous than others, they may be biased one way or the other, and they often have highly subjective preferences over the papers they review. Moreover, each reviewer usually has only a very local view, as he or she evaluates only a small fraction of the submissions. Despite all these shortcomings, the review scores obtained need to be aggregrated in order to globally rank all submissions and to make the acceptance/rejection decision. A common method is to simply take the average of each submission's review scores, possibly weighted by the reviewers' confidence levels. Unfortunately, the global ranking thus produced often suffers a certain unfairness, as the reviewers' biases and limitations are not taken into account. We propose a method for calibrating the scores of reviewers that are potentially biased and blindfolded by having only partial information. Our method uses a maximum likelihood estimator, which estimates both the bias of each individual reviewer and the unknown "ideal" score of each submission. This yields a quadratic program whose solution transforms the individual review scores into calibrated, globally comparable scores. We argue why our method results in a fairer and more reasonable global ranking than simply taking the average of scores. To show its usefulness, we test our method empirically using real-world data.