In past decade, more and more data are collected from multiple sources or represented by multiple views, where different views describe distinct perspectives of the data. Although each view could be individually used for finding patterns by clustering, the clustering performance could be more accurate by exploring the rich information among multiple views. Several multi-view clustering methods have been proposed to unsupervised integrate different views of data. However, they are graph based approaches, e.g. based on spectral clustering, such that they cannot handle the large-scale data. How to combine these heterogeneous features for unsupervised large-scale data clustering has become a challenging problem. In this paper, we propose a new robust large-scale multi-view clustering method to integrate heterogeneous representations of large-scale data. We evaluate the proposed new methods by six benchmark data sets and compared the performance with several commonly used clustering approaches as well as the baseline multi-view clustering methods. In all experimental results, our proposed methods consistently achieve superiors clustering performances.

Both SAT and #SAT can represent difficult problems in seemingly dissimilar areas such as planning, verification, and probabilistic inference. Here, we examine an expressive new language, #∃SAT, that generalizes both of these languages. #∃SAT problems require counting the number of satisfiable formulas in a concisely-describable set of existentially-quantified, propositional formulas. We characterize the expressiveness and worst-case difficulty of #∃SAT by proving that it is complete for the complexity class #P^{NP[1]} , and relating this class to more familiar complexity classes. We also experiment with three new general purpose #∃SAT solvers on a battery of problem distributions including a simple logistics domain. Our experiments show that, despite the formidable worst-case complexity of #P^{NP[1]} , many of the instances can be solved efficiently by noticing and exploiting a particular type of frequent structure.

In this paper, we consider the inclusion-exclusion rule – a known yet seldom used rule of probabilistic inference. Unlike the widely used sum rule which requires easy access to all joint probability values, the inclusion-exclusion rule requires easy access to several marginal probability values. We therefore develop a new representation of the joint distribution that is amenable to the inclusion-exclusion rule. We compare the relative strengths and weaknesses of the inclusion-exclusion rule with the sum rule and develop a hybrid rule called the inclusion- exclusion-sum (IES) rule, which combines their power. We apply the IES rule to junction trees, treating the latter as a target for knowledge compilation and show that in many cases it greatly reduces the time required to answer queries. Our experiments demonstrate the power of our approach. In particular, at query time, on several networks, our new scheme was an order of magnitude faster than the junction tree algorithm.

A key decision facing autonomous systems with access to streams of sensory data is whether to act based on current evidence or to wait for additional information that might enhance the utility of taking an action. Computing the value of information is particularly difficult with streaming high-dimensional sensory evidence. We describe a belief projection approach to reasoning about information value in these settings, using models for inferring future beliefs over states given streaming evidence. These belief projection models can be learned from data or constructed via direct assessment of parameters and they fit naturally in modular, hierarchical state inference architectures. We describe principles of using belief projection and present results drawn from an implementation of the methodology within a conversational system.

We have developed a method for using confidence scores to integrate labels provided by crowdsourcing workers. Although confidence scores can be useful information for estimating the quality of the provided labels, a way to effectively incorporate them into the integration process has not been established. Moreover, some workers are overconfident about the quality of their labels while others are underconfident, and some workers are quite accurate in judging the quality of their labels. This differing reliability of the confidence scores among workers means that the probability distributions for the reported confidence scores differ among workers. To address this problem, we extended the Dawid-Skene model and created two probabilistic models in which the values of unobserved true labels are inferred from the observed provided labels and reported confidence scores by using the expectation-maximization algorithm. Results of experiments using actual crowdsourced data for image labeling and binary question answering tasks showed that incorporating workers' confidence scores can improve the accuracy of integrated crowdsourced labels.

We study map-matching, the problem of estimating the route that is traveled by a vehicle, where the points observed with the Global Positioning System are available. A state-of-the-art approach for this problem is a Hidden Markov Model (HMM). We propose a particular transition probability between latent road segments by the use of the number of turns in addition to the travel distance between the latent road segments. We use inverse reinforcement learning to estimate the importance of the number of turns relative to the travel distance. This estimated importance is incorporated in the transition probability of the HMM. We show, through numerical experiments, that the error of map-matching can be reduced substantially with the proposed transition probability.

Probabilistic logics combine the expressive power of logic with the ability to reason with uncertainty. Several probabilistic logic languages have been proposed in the past, each of them with their own features. In this paper, we propose a new probabilistic constraint logic programming language, which combines constraint logic programming with probabilistic reasoning. The language supports modeling of discrete as well as continuous probability distributions by expressing constraints on random variables. We introduce the declarative semantics of this language, present an exact inference algorithm to derive bounds on the joint probability distributions consistent with the specified constraints, and give experimental results. The results obtained are encouraging, indicating that inference in our language is feasible for solving challenging problems.

This paper concerns building probabilistic models with an underlying ontology that defines the classes and properties used in the model. In particular, it considers the problem of reasoning with properties that may not always be defined. Furthermore, we may even be uncertain about whether a property is defined for a given individual. One approach is to explicitly add a value "undefined" to the range of random variables, forming extended belief networks; however, adding an extra value to a random variable's range has a large computational overhead. In this paper, we propose an alternative, ontologically-based belief networks, where all properties are only used when they are defined, and we show how probabilistic reasoning can be carried out without explicitly using the value "undefined" during inference. We prove this is equivalent to reasoning with the corresponding extended belief network and empirically demonstrate that inference becomes more efficient.

Parsing human poses in images is fundamental in extracting critical visual information for artificial intelligent agents. Our goal is to learn self-contained body part representations from images, which we call visual symbols, and their symbol-wise geometric contexts in this parsing process. Each symbol is individually learned by categorizing visual features leveraged by geometric information. In the categorization, we use Latent Support Vector Machine followed by an efficient cross validation procedure. Then, these symbols naturally define geometric contexts of body parts in a fine granularity for effective inference. When the structure of the compositional parts is a tree, we derive an efficient approach to estimating human poses in images. Experiments on two large datasets suggest our approach outperforms state of the art methods.

In this paper, we propose a novel method for object discovery and dense modelling in RGB-D image sequences using motion cues. We develop our method as a building block for active object perception, such that robots can learn about the environment through perceiving the effects of actions. Our approach simultaneously segments rigid-body motion within key views, and discovers objects and hierarchical relations between object parts. The poses of the key views are optimized in a graph of spatial relations to recover the rigid-body motion trajectories of the camera with respect to the objects. In experiments, we demonstrate that our approach finds moving objects, aligns partial views on the objects, and retrieves hierarchical relations between the objects.