In this paper, a novel method is developed for enabling Multi-Kernel Multi-Label Learning. Inter-label dependency and similarity diversity are simultaneously leveraged in the proposed method. A concept network is constructed to capture the inter-label correlations for classifier training. Maximal margin approach is used to effectively formulate the feature-label associations and the label-label correlations. Specific kernels are learned not only for each label but also for each pair of the inter-related labels. By learning the eigenfunctions of the kernels, the similarity between a new data point and the training samples can be computed in the online mode. Our experimental results on real datasets (web pages, images, music, and bioinformatics) have demonstrated the effectiveness of our method.

Many applications in Multilingual and Multimodal Information Access involve searching large databases of high dimensional data objects with multiple (conditionally independent) views. In this work we consider the problem of learning hash functions for similarity search across the views for such applications. We propose a principled method for learning a hash function for each view given a set of multiview training data objects. The hash functions map similar objects to similar codes across the views thus enabling cross-view similarity search. We present results from an extensive empirical study of the proposed approach which demonstrate its effectiveness on Japanese language People Search and Multilingual People Search problems.

We propose a general MCMC method for Bayesian inference in logic-based probabilistic modeling. It covers a broad class of generative models including Bayesian networks and PCFGs. The idea is to generalize an MCMC method for PCFGs to the one for a Turing-complete probabilistic modeling language PRISM in the context of statistical abduction where parse trees are replaced with explanations. We describe how to estimate the marginal probability of data from MCMC samples and how to perform Bayesian Viterbi inference using an example of Naive Bayes model augmented with a hidden variable.

Homophily and stochastic equivalence are two primary features of interest in social networks. Recently, the multiplicative latent factor model (MLFM) is proposed to model social networks with directed links. Although MLFM can capture stochastic equivalence, it cannot model well homophily in networks. However, many real-world networks exhibit homophily or both homophily and stochastic equivalence, and hence the network structure of these networks cannot be modeled well by MLFM. In this paper, we propose a novel model, called generalized latent factor model (GLFM), for social network analysis by enhancing homophily modeling in MLFM. We devise a minorization-maximization (MM) algorithm with linear-time complexity and convergence guarantee to learn the model parameters. Extensive experiments on some real-world networks show that GLFM can effectively model homophily to dramatically outperform state-of-the-art methods.

Transfer learning addresses the problems that labeled training data are insufficient to produce a high-performance model. Typically, given a target learning task, most transfer learning approaches require to select one or more auxiliary tasks as sources by the designers. However, how to select the right source data to enable effective knowledge transfer automatically is still an unsolved problem, which limits the applicability of transfer learning. In this paper, we take one step ahead and propose a novel transfer learning framework, known as source-selection-free transfer learning (SSFTL), to free users from the need to select source domains. Instead of asking the users for source and target data pairs, as traditional transfer learning does, SSFTL turns to some online information sources such as World Wide Web or the Wikipedia for help. The source data for transfer learning can be hidden somewhere within this large online information source, but the users do not know where they are. Based on the online information sources, we train a large number of classifiers. Then, given a target task, a bridge is built for labels of the potential source candidates and the target domain data in SSFTL via some large online social media with tag cloud as a label translator. An added advantage of SSFTL is that, unlike many previous transfer learning approaches, which are difficult to scale up to the Web scale, SSFTL is highly scalable and can offset much of the training work to offline stage. We demonstrate the effectiveness and efficiency of SSFTL through extensive experiments on several real-world datasets in text classification.

We propose a method to predict users’ interests in social media, using time-evolving, multinomial relational data. We exploit various actions performed by users, and their preferences to predict user interests. Actions performed by users in social media such as Twitter, Delicious and Facebook have two fundamental properties. (a) User actions can be represented as high-dimensional or multinomial relations - e.g. referring URLs, bookmarking and tagging, clicking a favorite button on a post etc. (b) User actions are time-varying and user-specific – each user has unique preferences that change over time. Consequently, it is appropriate to represent each user’s action at some point in time as a multinomial relational data. We propose ActionGraph, a novel graph representation for modeling users’ multinomial, time-varying actions. Each user’s action at some time point is represented by an action node. ActionGraph is a bipartite graph whose edges connect an action node to its involving entities, referred to as object nodes. Using real-world social media data, we empirically justify the proposed graph structure. Our experimental results show that the proposed ActionGraph improves the accuracy in a user interest prediction task by outperforming several baselines including standard tensor analysis, a previously proposed state-of-the-art LDA-based method and other graph-based variants. Moreover, the proposed method shows robust performances in the presence of sparse data.

Activity recognition aims to discover one or more users’ actions and goals based on sensor readings. In the real world, a single user’s data are often insufficient for training an activity recognition model due to the data sparsity problem. This is especially true when we are interested in obtaining a personalized model. In this paper, we study how to collaboratively use different users’ sensor data to train a model that can provide personalized activity recognition for each user. We propose a user-dependent aspect model for this collaborative activity recognition task. Our model introduces user aspect variables to capture the user grouping information, so that a target user can also benefit from her similar users in the same group to train the recognition model. In this way, we can greatly reduce the need for much valuable and expensive labeled data required in training the recognition model for each user. Our model is also capable of incorporating time information and handling new user in activity recognition. We evaluate our model on a real-world WiFi data set obtained from an indoor environment, and show that the proposed model can outperform several state-of-art baseline algorithms.

Occlusions are a central phenomenon in multi-object computer vision. However, formal analyses (LOS14, ROC20) proposed in the spatial reasoning literature ignore many distinctions crucial to computer vision, as a result of which these algebras have been largely ignored in vision applications. Two distinctions of relevance to visual computation are (a) whether the occluder is a moving object or part of the static background, and (b) whether the visible part of an object is a connected blob or fragmented. In this work, we develop a formal model of occlusion states that combines these criteria with overlap distinctions modeled in spatial reasoning to come up with a comprehensive set of fourteen occlusion states, which we define as OCS14. Transitions between these occlusion states are an important source of information on visual activity (e.g. splits and merges). We show that the resulting formalism is representationally complete in the sense that these states constitute a partition of all possible occlusion situations based on these criteria. Finally, we show results from implementations of this approach in a test application involving static camera based scene analysis, where occlusion state analysis and multiple object tracking can be used for two tasks -- (a) identifying static occluders, and (b) modeling a class of interactions represented as transitions of occlusion states. Thus, the formalism is shown to have direct relevance to actual vision applications.

Forming effective coalitions is a major research challenge in AI and multi-agent systems (MAS). Thus, coalitional games, including Coalition Structure Generation (CSG), have been attracting considerable attention from the AI research community. Traditionally, the input of a coalitional game is a black-box function called a characteristic function. A range of previous studies have found that many problems in coalitional games tend to be computationally intractable when the input is a black-box function. Recently, several concise representation schemes for a characteristic function have been proposed. Although these schemes are effective for reducing the representation size, most problems remain computationally intractable. In this paper, we develop a new concise representation scheme based on the idea of agent types. Intuitively, a type represents a set of agents, which are recognized as having the same contribution. This representation can be exponentially more concise than existing concise representation schemes. Furthermore, this idea can be used in conjunction with existing schemes to further reduce the representation size. Moreover, we show that most of the problems in coalitional games, including CSG, can be solved in polynomial time in the number of agents, assuming the number of possible types is fixed.

Most of the text mining tasks, such as clustering, is dominated by statistical approaches that treat text as a bag of words. Semantics in the text is largely ignored in the mining process, and the mining results are often not easily interpretable. One particular challenge faced by such approaches is short text understanding, as short text lacks enough content from which a statistical conclusion can be drawn. For example, traditional topic analysis methods consider topic segments with tens of hundreds of words. Latent topic modeling, such as latent Dirichlet allocation, also requires sufficient words to infer document topic distribution. We enhance machine learning algorithms by first giving the machine a probabilistic knowledgebase that contains as big, rich, and consistent concepts (of worldly facts) as those in our mental world. Then a Bayesian inference mechanism is developed to conceptualize words and short text. We conducted comprehensive tests of our method on conceptualizing set of text terms, as well as clustering Twitter messages (tweets), which are typically approximately ten words long. Compared to latent semantic topic modeling and other four kinds of methods that using WordNet, Freebase and Wikipedia (category links and explicit semantic analysis), we show significant improvements in terms of tweets clustering accuracy.