Multi-modal data is dramatically increasing with the fast growth of social media. Learning a good distance measure for data with multiple modalities is of vital importance for many applications, including retrieval, clustering, classification and recommendation. In this paper, we propose an effective and scalable multi-modal distance metric learning framework. Based on the multi-wing harmonium model, our method provides a principled way to embed data of arbitrary modalities into a single latent space, of which an optimal distance metric can be learned under proper supervision, i.e., by minimizing the distance between similar pairs whereas maximizing the distance between dissimilar pairs. The parameters are learned by jointly optimizing the data likelihood under the latent space model and the loss induced by distance supervision, thereby our method seeks a balance between explaining the data and providing an effective distance metric, which naturally avoids overfitting. We apply our general framework to text/image data and present empirical results on retrieval and classification to demonstrate the effectiveness and scalability.

People’s facial expressions, whether made consciously or subconsciously, continuously revealtheir state of mind. This work proposes a methodfor predicting people’s strategic decisions based ontheir facial expressions. We designed a new version of the centipede game that intorduces an incentive for the human participant to hide her facial expressions. We recorded on video participants whoplayed several games of our centipede version, andconcurrently logged their decisions throughout thegames. The video snippet of the participants’ facesprior to their decisions is represented as a fixed-size vector by estimating the covariance matrix of keyfacial points which change over time. This vectorserves as input to a classifier that is trained to predict the participant’s decision. We compare severaltraining techniques, all of which are designed towork with the imbalanced decisions typically madeby the players of the game. Furthermore, we investigate adaptation of the trained model to eachplayer individually, while taking into account theplayer’s facial expressions in the previous games.The results show that our method outperforms standard SVM as well as humans in predicting subjects’strategic decisions. To the best of our knowledge,this is the first study to present a methodology forpredicting people’s strategic decisions when thereis an incentive to hide facial expressions.

Eliciting user preferences constitutes a major step towards developing recommender systems and decision support tools. Assuming that preferences are ceteris paribus allows for their concise representation as Conditional Preference Networks (CP-nets). This work presents the first empirical investigation of an algorithm for reliably and efficiently learning CP-nets in a manner that is minimally intrusive . At the same time, it introduces a novel process for efficiently reasoning with (the learned) preferences.

We analyze the foundations of cyclic causal models for discrete variables, and compare structural equation models (SEMs) to an alternative semantics as the equilibrium (stationary) distribution of a Markov chain. We show under general conditions, discrete cyclic SEMs cannot have independent noise; even in the simplest case, cyclic structural equation models imply constraints on the noise. We give a formalization of an alternative Markov chain equilibrium semantics which requires not only the causal graph, but also a sample order. We show how the resulting equilibrium is a function of the sample ordering, both theoretically and empirically.

We learn constraint networks by asking the user partial queries. That is, we ask the user to classify assignments to subsets of the variables as positive or negative. We provide an algorithm that, given a negative example, focuses onto a constraint of the target network in a number of queries logarithmic in the size of the example. We give information theoretic lower bounds for learning some simple classes of constraint networks and show that our generic algorithm is optimal in some cases. Finally we evaluate our algorithm on some benchmarks.

This paper proposes a new algorithm to compute the resilience of a social system or an ecosystem when it is defined in the framework of the mathematical viability theory. It is applied to the problem of language coexistence: Although bilingual societies do exist, many languages have disappeared and some seem endangered presently. Mathematical models of language competition generally conclude that one language will disappear, except when the relative prestige of the languages can be modified. The viability theory provides concepts and tools that are suitable to study the resilience, but with severe computational limits since it uses extensive search on regular grids. The method we propose considers the computation of the viability output sets as an active learning problem with the objective of restraining the number of calls to the model and information storage. We adapt a kd-tree algorithm to approximate the level sets of the resilience value. We prove that this algorithm converges to the output sets defined by the viability theory (viability kernel and capture basin). The resilience value we compute can then be used to propose a policy of action in risky situations such as migration flows.

Real-world problems generally involve several antagonistic objectives, like quality and cost for design problems, or makespan and cost for planning problems. The only approaches to multiobjective AI Planning rely on metrics, that can incorporate several objectives in some linear combinations, and metric sensitive planners, that are able to give different plans for different metrics, and hence to eventually approximate the Pareto front of the multiobjective problem, i.e. the set of optimal trade-offs between the antagonistic objectives. Divide-and-Evolve (DaE) is an evolutionary planner that embeds a classical planner and feeds it with a sequence of subproblems of the problem at hand. Like all Evolutionary Algorithms, DaE can be turned into a Pareto-based multiobjective solver, even though using an embedded planner that is not metric sensitive. The Pareto-based multiobjective planner MO-DaE thus avoids the drawbacks of the aggregation method. Furthermore, using YAHSP as the embedded planner, it outperforms in many cases the metric-based approach using LPG metric sensitive planner, as witnessed by experimental results on original multiobjective benchmarks built upon IPC-2011 domains.

This paper presents a new application of logic programming to a real-life problem in hydraulic engineering. The work is developed as a collaboration of computer scientists and hydraulic engineers, and applies Constraint Logic Programming to solve a hard combinatorial problem. This application deals with one aspect of the design of a water distribution network, i.e., the valve isolation system design. We take the formulation of the problem by [Giustolisi and Savic, 2008] and show how, thanks to constraint propagation, we can get better solutions than the best solution known in the literature for the Apulian distribution network.

When using graphical models for decision making, the presence of unobserved variables  may hinder our ability to reach the correct decision.  A fundamental question here is whether or not  one is ready to make a decision (stopping criteria), and if not, what additional  observations should be made in order to better prepare for a decision  (selection criteria).  A recently introduced notion, the Same-Decision Probability (SDP), has been shown to be useful  as both a stopping and a selection criteria.  This query has been shown to be highly intractable,  being PP^PP-complete, and is exemplary of a class of queries which correspond to the computation of certain expectations.  We propose the first exact algorithm for computing the SDP in this paper, and demonstrate its effectiveness on several real and synthetic networks.  We also present a new complexity result for computing the SDP on models with a Naive Bayes structure.

Multi-level logic synthesis is a problem of immense practical significance, and is a key to developing circuits that optimize a number of parameters, such as depth, energy dissipation, reliability, etc. The problem can be defined as the task of taking a collection of components from which one wants to synthesize a circuit that optimizes a particular objective function. This problem is computationally hard, and there are very few automated approaches for its solution. To solve this problem we propose an algorithm, called Circuit-Decomposition Engine (CDE), that is based on learning decision trees, and uses a greedy approach for function learning. We empirically demonstrate that CDE, when given a library of different component types, can learn the function of Disjunctive Normal Form (DNF) Boolean representations and synthesize circuit structure using the input library. We compare the structure of the synthesized circuits with that of well-known circuits using a range of circuit similarity metrics.