Evaluating an agent's performance in a stochastic setting is necessary for agent development, scientific evaluation, and competitions. Traditionally, evaluation is done using Monte Carlo estimation; the magnitude of the stochasticity in the domain or the high cost of sampling, however, can often prevent the approach from resulting in statistically significant conclusions. Recently, an advantage sum technique has been proposed for constructing unbiased, low variance estimates of agent performance. The technique requires an expert to define a value function over states of the system, essentially a guess of the state's unknown value. In this work, we propose learning this value function from past interactions between agents in some target population. Our learned value functions have two key advantages: they can be applied in domains where no expert value function is available and they can result in tuned evaluation for a specific population of agents (e.g., novice versus advanced agents). We demonstrate these two advantages in the domain of poker. We show that we can reduce variance over state-of-the-art estimators for a specific population of limit poker players as well as construct the first variance reducing estimators for no-limit poker and multi-player limit poker.

Manifold alignment has been found to be useful in many areas of machine learning and data mining. In this paper we introduce a novel manifold alignment approach, which differs from semi-supervised alignment and Procrustes alignment in that it does not require predetermining correspondences. Our approach learns a projection that maps data instances (from two different spaces) to a lower dimensional space simultaneously matching the local geometry and preserving the neighborhood relationship within each set. This approach also builds connections between spaces defined by different features and makes direct knowledge transfer possible. The performance of our algorithm is demonstrated and validated in a series of carefully designed experiments in information retrieval and bioinformatics.

This paper deals with evolutionary clustering, which refers to the problem of clustering data with distribution drifting along time. Starting from a density estimation view to clustering problems, we propose two general on-line frameworks. In the first framework, i.e., historical data dependent (HDD), current model distribution is designed to approximate both current and historical data distributions. In the second framework, i.e., historical model dependent (HMD), current model distribution is designed to approximate both current data distribution and historical model distribution. Both frameworks are based on the general exponential family mixture (EFM) model. As a result, all conventional clustering algorithms based on EFMs can be extended to evolutionary setting under the two frameworks. Empirical results validate the two frameworks.

Question-answer pairs extracted from email threads can help construct summaries of the thread, as well as inform semantic-based assistance with email. Previous work dedicated to email threads extracts only questions in interrogative form. We extend the scope of question and answer detection and pairing to encompass also questions in imperative and declarative forms, and to operate at sentence-level fidelity. Building on prior work, our methods are based on learned models over a set of features that include the content, context, and structure of email threads. For two large email corpora, we show that our methods balance precision and recall in extracting question-answer pairs, while maintaining a modest computation time.

Most of the existing metric learning methods are accomplished by exploiting pairwise constraints over the labeled data and frequently suffer from the insufficiency of training examples.  To learn a robust distance metric from few labeled examples, prior knowledge from unlabeled examples as well as the metrics previously derived from auxiliary data sets can be useful.  In this paper, we propose to leverage such auxiliary knowledge to assist distance metric learning, which is formulated following the regularized loss minimization principle.  Two algorithms are derived on the basis of manifold regularization and log-determinant divergence regularization technique, respectively, which can simultaneously exploit label information (i.e., the pairwise constraints over labeled data), unlabeled examples, and the metrics derived from auxiliary data sets.  The proposed methods directly manipulate the auxiliary metrics and require no raw examples from the auxiliary data sets, which make them efficient and flexible.  We conduct extensive evaluations to compare our approaches with a number of competing approaches on face recognition task.  The experimental results show that our approaches can derive reliable distance metrics from limited training examples and thus are superior in terms of accuracy and labeling efforts.

This study examines the ability of a semantic space model to represent the meaning of noun compounds such as "information gathering" or "weather forecast," A new algorithm,  comparison, is proposed for computing compound vectors from constituent word vectors, and compared with other algorithms (i.e., predication and centroid) in terms of accuracy of multiple-choice synonym test and similarity judgment test. The result of both tests is that the comparison algorithm is, on the whole, superior to other algorithms, and in particular achieves the best performance when noun compounds have emergent meanings. Furthermore, the comparison algorithm also works for novel noun compounds that do not occur in the corpus. These findings indicate that a semantic space model in general and the comparison algorithm in particular has sufficient ability to compute the meaning of noun compounds.

As data-center energy consumption continues to rise, efficient power management is becoming increasingly important. In this work, we examine the use of a novel market mechanism for finding the right balance between power and performance. The market enables a separation between a `buyer side' that strives to maximize performance and a 'seller side' that strives to minimize power and other costs. A concise and scalable description language is defined for agent preferences that admits a mixed-integer program for computing optimal allocations. Experimental results demonstrate the robustness, flexibility, practicality and scalability of the architecture.

But this premature paving is not very useful if the searchtree is explored depth-first (DFS) or breadth-first (BFS): DFS When applied to numerical CSPs, the branch and quickly converges to ɛ-boxes that are too close to one another prune algorithm (BPA) computes a sharp covering to be representative of the solution set (see the left part of of the solution set. The BPA is therefore impractical Figure 1); BFS computes a homogeneous paving but finds no when the solution set is large, typically when ɛ-box at all if stopped too early (see the center graphic of Figure it has a dimension larger than four or five which is 1; note that such a sharp paving cannot be computed for often met in underconstrained problems. The purpose larger solution sets, making BFS useless in such cases). of this paper is to present a new search tree The search strategy used in an anytime BPA should quickly exploration strategy for BPA that hybridizes depthfirst find ɛ-boxes that are representative of the solution set: ɛ- and breadth-first searches. This search strategy boxes should be discovered uniformly on a continuous connected allows the BPA discovering potential solutions component in the solution set, while every connected in different areas of the search space in early stages components should be reached by some ɛ-boxes in early of the exploration, hence allowing an anytime usage stages of the search. Two such strategies are introduced in of the BPA. The merits of the proposed search the present paper. The most distant-first strategy (MDFS) strategy are experimentally evaluated.

We then describe In this paper, we describe a user evaluation of a humanrobot a study which assessed the responses of naïve users dialogue system that is designed to enable a humanoid to output that varied along two dimensions: the robot to cooperate with a human partner on building wooden method of describing an assembly plan (pre-order construction toys. In the evaluation, we experimentally vary or post-order), and the method of referring to objects two aspects of the output generated by the system: the way in the world (basic and full). Varying both that it describes assembly plans to the user, and the way that of these factors produced significant results: subjects it refers to objects in the world. We then measure the impact using the system that employed a pre-order of varying each of these features on the users' objective success description strategy asked for instructions to be repeated at working with the system, as well as on their subjective significantly less often than those who experienced impressions of the interaction.

Splitting a logic program allows us to reduce the task of computing its stable models to similar tasks for smaller programs.  This idea is extended here to the general theory of stable models that replaces traditional logic programs by arbitrary first-order sentences and distinguishes between intensional and extensional predicates.  We discuss two kinds of splitting: a set of intensional predicates can be split into subsets, and a formula can be split into its conjunctive terms.