We consider the high-dimensional heteroscedastic regression model, where the mean and the log variance are modeled as a linear combination of input variables. Existing literature on high-dimensional linear regres- sion models has largely ignored non-constant error variances, even though they commonly occur in a variety of applications ranging from biostatis- tics to finance. In this paper we study a class of non-convex penalized pseudolikelihood estimators for both the mean and variance parameters. We show that the Heteroscedastic Iterative Penalized Pseudolikelihood Optimizer (HIPPO) achieves the oracle property, that is, we prove that the rates of convergence are the same as if the true model was known. We demonstrate numerical properties of the procedure on a simulation study and real world data.

We provide a unifying framework linking two classes of statistics used in two-sample and independence testing: on the one hand, the energy distances and distance covariances from the statistics literature; on the other, distances between embeddings of distributions to reproducing kernel Hilbert spaces (RKHS), as established in machine learning. The equivalence holds when energy distances are computed with semimetrics of negative type, in which case a kernel may be defined such that the RKHS distance between distributions corresponds exactly to the energy distance. We determine the class of probability distributions for which kernels induced by semimetrics are characteristic (that is, for which embeddings of the distributions to an RKHS are injective). Finally, we investigate the performance of this family of kernels in two-sample and independence tests: we show in particular that the energy distance most commonly employed in statistics is just one member of a parametric family of kernels, and that other choices from this family can yield more powerful tests.

In this paper, we outline a system for evaluating the performance of scientific research across a number of outcome metrics (e.g. publications, sales, new hires). Our system is designed to classify research performance into a number of metrics, evaluate each metric’s performance using only data on other metrics, and to cast predictions of future performance by metric. This study shows how data mining techniques can be used to provide a predictive analytic approach to the management of resources for scientific research.

The formal semantics of speech acts, even in the classical framework of illocutionary logic, requires considerations that go beyond individual speech activity and beyond the interpretation of individual sentences. We show how the formal semantics of speech acts can be extended to take into account the social effects and interactive aspects of illocutionary activity. To illustrate our approach, we focus on the semantics of assertions and descriptive discourse, contrasting the individual aspect of speaker's meaning and the epistemic effects of assertion making. The approach presented in this paper generalizes to all other types of illocutionary acts, adding specific content to the conversational record that registers the common ground of speakers and hearers as a dialogue unfolds.

Household appliances such as dishwashers, televisions and radios are an indispensable part of the modern household. Yet, people who have some form of physical impairment often find that they are unable to make use of these commonly available appliances, to the detriment of their lifestyle. This paper proposes a gesture interface for home appliances that can be used by people with physical impairments. Two simulated gesture controlled appliances are developed and evaluated by physically impaired people. The results show that this interface is able to allow physically impaired people to make use of modern appliances by gesture.

Reading tutors can provide an unprecedented opportunity to collect and analyze large amounts of data for understanding how students learn. We trained models of oral reading prosody (pitch, intensity, and duration) on a corpus of narrations of 4558 sentences by 11 fluent adults. We used these models to evaluate the oral reading prosody of 85,209 sentences read by 55 children (mostly) 7-10 years old who used Project LISTEN's Reading Tutor during the 2005-2006 school year. We mined the resulting data to pinpoint the specific common syntactic and lexical features of text that children scored best and worst on. These features predict their fluency and comprehension test scores and gains better than previous models. Focusing on these features may help human or automated tutors improve children’s fluency and comprehension more effectively.

Artifact capabilities can play an important role in understanding human cognition. Over time humans learn to use artifacts, evolve the knowledge and combine acquired capabilities with others to form complex capabilities. In this study we present a hybrid ontology of artifacts to facilitate learning artifact capabilities. We develop a framework where agents simultaneously exploit a centralized artifact ontology in the environment and a distributed artifact ontology local to each agent. We demonstrate how both ontologies can be used by agents both in the artifact selection process and in learning artifact use. The local ontology serves as domain knowledge gained by the agent as it learns. We illustrate an example to show how an acquired artifact capability can be stored in an agent's local ontology for future use.

The quality criteria of system P have been guiding qualitative uncertain reasoning now for more than two decades. Different semantical approaches have been presented to provide semantics for system P. The aim of the present paper is to investigate the semantical structures underlying system P in more detail, namely, on the level of the models. In particular, we focus on the approach via conditional objects which relies on Boolean intervals, without making any use of qualitative or quantitative information. Indeed, our studies confirm the singular position of conditional objects, but we are also able to establish semantical relationships via novel variants of model theories.

In this paper, we present a programming paradigm for AI problem solving based on computational concepts drawn from Affective Computing. It is believed that emotions participate in human adaptability and reactivity, in behaviour selection and in complex and dynamic environments. We propose to define a mechanism inspired from this observation for general AI problem solving. To this purpose, we synthesize emotions as programming abstractions that represent the perception of the environment's state w.r.t. predefined heuristics such as goal distance, action capability,etc. We first describe the general architecture of this "emotion-oriented" programming model. We define the vocabulary that allows programmers to describe the problem to be solved (i.e. the environment), and the action selection function based on emotion abstractions (i.e. the agent's behaviours). We then present the runtime algorithm that builds emotions out of the environment, stores them in the agent's memory, and selects behaviours accordingly. We present the implementation of a classical labyrinth problem solver in this model. We show that the solutions obtained by this easy-to-implement emotion-oriented program are of good quality while having a reduced computational cost.

Several authors analyzed propositional attitudes ( wish , fear , regret , glad ) by integrating their epistemic and deontic components. This paper extends previous work done by the author and presents a logical calculus inspired by Possibility Theory, a non-compositional version of fuzzy logic.