Sparse coding is an unsupervised learning algorithm for finding concise, slightly higher-level representations of inputs, and has been successfully applied to self-taught learning, where the goal is to use unlabeled data to help on a supervised learning task, even if the unlabeled data cannot be associated with the labels of the supervised task (Raina et al., 2007). However, sparse coding uses a Gaussian noise model and a quadratic loss function, and thus performs poorly if applied to binary valued, integer valued, or other non-Gaussian data, such as text. Drawing on ideas from generalized linear models (GLMs), we present a generalization of sparse coding to learning with data drawn from any exponential family distribution (such as Bernoulli, Poisson, etc). This gives a method that we argue is much better suited to model other data types than Gaussian. We present an algorithm for solving the L1-regularized optimization problem defined by this model, and show that it is especially efficient when the optimal solution is sparse. We also show that the new model results in significantly improved self-taught learning performance when applied to text classification and to a robotic perception task.

We present an algorithm for selecting an appropriate abstraction when learning a new skill. We show empirically that it can consistently select an appropriate abstraction using very little sample data, and that it significantly improves skill learning performance in a reasonably large real-valued reinforcement learning domain.

Consider the setting where a panel of judges is repeatedly asked to (partially) rank sets of objects according to given criteria, and assume that the judges' expertise depends on the objects' domain.  Learning to aggregate their rankings with the goal of producing a better joint ranking is a fundamental problem in many areas of Information Retrieval and Natural Language Processing, amongst others.  However, supervised ranking data is generally difficult to obtain, especially if coming from multiple domains.  Therefore, we propose a framework for learning to aggregate votes of constituent rankers with domain specific expertise without supervision.  We apply the learning framework to the settings of aggregating full rankings and aggregating top-k lists, demonstrating significant improvements over a domain-agnostic baseline in both cases.

Local pattern mining is concerned with finding the set of patterns that satisfy a constraint in a database. We study local pattern mining in the context of ProbLog, a probabilistic Prolog system, and introduce an approach for finding correlated patterns in the form of queries in such a Prolog system. The approach combines principles of inductive logic programming, data mining and statistical relational learning. Experiments on a challenging biological network mining task provide evidence for the interestingness of the approach.

In this work we propose gRegress, a new algorithm which given set of labeled graphs and a real value associated with each graph extracts the complete set of subgraphs such that a) each subgraph in this set has correlation with the real value above a user-specified threshold and b) each subgraph in this set has correlation with any other subgraph in the set below a user-specified threshold. gRegress incorporates novel pruning mechanisms based on correlation of a subgraph feature with the output and correlation with other subgraph features. These pruning mechanisms lead to significant speedup. Experimental results indicate that in terms of runtime, gRegress substantially outperforms gSpan, often by an order of magnitude while the regression models produced by both approaches have comparable accuracy.

In this paper, we consider semi-supervised classification on evolutionary data, where the distribution of the data and the underlying concept that we aim to learn change over time due to short-term noises and long-term drifting, making a single aggregated classifier inapplicable for long-term classification. The drift is smooth if we take a localized view over the time dimension, which enables us to impose temporal smoothness assumption for the learning algorithm. We first discuss how to carry out such assumption using temporal regularizers defined in a structural way with respect to the Hilbert space, and then derive the online algorithm that efficiently finds the closed-form solution to the classification functions. Experimental results on real-world evolutionary mailing list data demonstrate that our algorithm outperforms classical semi-supervised learning algorithms in both algorithmic stability and classification accuracy.

Dimensionality reduction is an essential step in high-dimensional data analysis. Many dimensionality reduction algorithms have been applied successfully to multi-class and multi-label problems. They are commonly applied as a separate data preprocessing step before classification algorithms. In this paper, we study a joint learning framework in which we perform dimensionality reduction and multi-label classification simultaneously. We show that when the least squares loss is used in classification, this joint learning decouples into two separate components, i.e., dimensionality reduction followed by multi-label classification. This analysis partially justifies the current practice of a separate application of dimensionality reduction for classification problems. We extend our analysis using other loss functions, including the hinge loss and the squared hinge loss. We further extend the formulation to the more general case where the input data for different class labels may differ, overcoming the limitation of traditional dimensionality reduction algorithms. Experiments on benchmark data sets have been conducted to evaluate the proposed joint formulations.

Bootstrap Voting Experts (BVE) is an extension to the Voting Experts algorithm for unsupervised chunking of sequences. BVE generates a series of segmentations, each of which incorporates knowledge gained from the previous segmentation. We show that this method of bootstrapping improves the performance of Voting Experts in a variety of unsupervised word segmentation scenarios, and generally improves both precision and recall of the algorithm. We also show that Minimum Description Length (MDL) can be used to choose nearly optimal parameters for Voting Experts in an unsupervised manner.

Recently graph based dimensionality reduction has received a lot of interests in many fields of information processing. Central to it is a graph structure which models the geometrical and discriminant structure of the data manifold. When label information is available, it is usually incorporated into the graph structure by modifying the weights between data points. In this paper, we propose a novel dimensionality reduction algorithm, called Constrained Graph Embedding, which considers the label information as additional constraints. Specifically, we constrain the space of the solutions that we explore only to contain embedding results that are consistent with the labels. Experimental results on two real life data sets illustrate the effectiveness of our proposed method.

We propose an approach for automatically ranking structured documents applied to patent prior art search. Our model, SVM Patent Ranking (SVM_PR) incorporates margin constraints that directly capture the specificities of patent citation ranking. Our approach combines patent domain knowledge features with meta-score features from several different general Information Retrieval methods. The training algorithm is an extension of the Pegasos algorithm with performance guarantees, effectively handling hundreds of thousands of patent-pair judgements in a high dimensional feature space. Experiments on a homogeneous essential wireless patent dataset show that SVM_PR performs on average 30%-40% better than many other state-of-the-art general-purpose Information Retrieval methods in terms of the NDCG measure at different cut-off positions.