In this work, we make a contribution to developing turn-taking mechanism in spoken dialogue systems. We focus on modelling the turn-taking behavior in human-human conversations. The proposed models are tested on the Switchboard corpus which contains conversations annotated at the utterance level. Several experiments were performed to analyze the salience of different features that are associated with the preceding utterances for the task of predicting whether there will be a change in speaker. The impact of the n-gram sequential modelling on turn-taking is studied. Machine learning techniques are also employed to perform this prediction task. Results from the experiments suggest that a combination of the preceding dialogue sequence, previous changes in speaker information and duplicating the sequences by replacing speaker IDs plays an important role in modelling turn-taking. Utterance sequences of length 3 in N-grams resulted in higher predictability for this task. Experiments suggest that a machine learning technique with 4-grams of a combination of all these features is effective for predicting speaker changes.

The big challenge for Artificial Intelligence is a better understanding of human nature. Our fundamental motivation is to understand the minds of modern people by uncovering mechanisms of the brain, genes, and body, and enhancing our health and cognitive talents with Artificial Intelligence technologies. This paper presents how we can quantify cognitive biases in the decision-making process and understand the evolutionary mechanisms using Ambient Intelligence and crowdsourced genetics technologies. We focus on prospect theory (proposed by Daniel Kahneman), which models how people choose between options involving gains or losses. People perceive losses to hurt more than gains feel good. This “loss aversion” is an important cognitive bias in decision-making. However, little is known about individual differences in loss aversion. We launched a citizen science project to test the hypothesis that mutations in genes related to neural processes are related to individual variation in loss aversion. Our preliminary experiment showed that DRD2 gene mutations may be related to individual variation in loss aversion. This crowdsourced genetics research is probably the first trial to report the possibilities of individual genetic differences in loss aversion behaviors. We discuss the future paradigms in Ambient Intelligence for health and cognitive enhancement.

We introduce the concept of Hypoelliptic Diffusion Maps (HDM), a framework generalizing Diffusion Maps in the context of manifold learning and dimensionality reduction. Standard non-linear dimensionality reduction methods (e.g., LLE, ISOMAP, Laplacian Eigenmaps, Diffusion Maps) focus on mining massive data sets using weighted affinity graphs; Orientable Diffusion Maps and Vector Diffusion Maps enrich these graphs by attaching to each node also some local geometry. HDM likewise considers a scenario where each node possesses additional structure, which is now itself of interest to investigate. Virtually, HDM augments the original data set with attached structures, and provides tools for studying and organizing the augmented ensemble. The goal is to obtain information on individual structures attached to the nodes and on the relationship between structures attached to nearby nodes, so as to study the underlying manifold from which the nodes are sampled. In this paper, we analyze HDM on tangent bundles, revealing its intimate connection with sub-Riemannian geometry and a family of hypoelliptic differential operators. In a later paper, we shall consider more general fibre bundles.

Latent tree graphical models are a popular class of latent variable models, where a probability distribution involving observed and hidden variables are Markovian on a tree. Due to the fact that structure of (observable and hidden) variable interactions are approximated as a tree, inference on latent trees can be carried out exactly through a simple belief propagation [Pea88]. Therefore, latent tree graphical models present a good tradeoff between model accuracy and computational complexity. They are applicable in many domains, where it is natural to expect hierarchical or sequential relationships among the variables (through a hidden-Markov model). For instance, latent tree models have been employed for phylogenetic reconstruction [DEKM99], object recognition [CTW12a, CTW12b] and human pose estimation [WL13]. In this paper, we use latent tree model for discovering a hierarchy among diseases based on comorbidities exhibited in patients' health records, i.e. co-occurrences of diseases in patients. In particular, two large healthcare datasets of 30K and 1.6M patients are used to build the latent disease trees, where clinically meaningful disease clusters are identified as shown in fig 3 and 4. The task of learning a latent tree models consists of two parts: learning the tree structure, and learning the parameters of the tree. There exist many challenges which prohibit efficient or guaranteed learning of the latent tree graphical model, which will be addressed in this paper: 1. The location and the number of latent variables are hidden and the marginalized graph over the observable variables no longer conforms to a tree structure.

Sequential compressed sensing is a promising new information acquisition and recovery technique to process big data that arise in various applications such as compressive imaging [2]-[4], power network monitoring [5], and large scale sensor networks [6]. The sequential nature of the problems arises either because the measurements are taken one after another, or due to the fact that the data is obtained in a streaming fashion so that it has to be processed in one pass. To harvest the benefits of adaptivity in sequential compressed sensing, various algorithms have been developed (see [1] for a review.) We may classify these algorithms as (1) being agnostic about the signal distribution and, hence, using random measurements [7]-[13]; (2) exploiting additional structure of the signal (such as graphical structure [14] and tree-sparse structure [15], [16]) to design measurements; (3) exploiting the distributional information of the signal in choosing the measurements possibly through maximizing mutual information: the seminal Bayesian compressive sensing work [17], Gaussian mixture models (GMM) [18], [19] and our earlier work [1] which presents a general framework for information guided sensing referred to as Info-Greedy Sensing. In this paper we consider the setup of Info-Greedy Sensing [1], as it provides certain optimality guarantees. Info-Greedy Sensing aims at designing subsequent measurements to maximize the mutual information conditioned on previous measurements.

Active learning is a type of sequential design for supervised machine learning, in which the learning algorithm sequentially requests the labels of selected instances from a large pool of unlabeled data points. The objective is to produce a classifier of relatively low risk, as measured under the 0-1 loss, ideally using fewer label requests than the number of random labeled data points sufficient to achieve the same. This work investigates the potential uses of surrogate loss functions in the context of active learning. Specifically, it presents an active learning algorithm based on an arbitrary classification-calibrated surrogate loss function, along with an analysis of the number of label requests sufficient for the classifier returned by the algorithm to achieve a given risk under the 0-1 loss. Interestingly, these results cannot be obtained by simply optimizing the surrogate risk via active learning to an extent sufficient to provide a guarantee on the 0-1 loss, as is common practice in the analysis of surrogate losses for passive learning. Some of the results have additional implications for the use of surrogate losses in passive learning.

Many high dimensional sparse learning problems are formulated as nonconvex optimization. A popular approach to solve these nonconvex optimization problems is through convex relaxations such as linear and semidefinite programming. In this paper, we study the statistical limits of convex relaxations. Particularly, we consider two problems: Mean estimation for sparse principal submatrix and edge probability estimation for stochastic block model. We exploit the sum-of-squares relaxation hierarchy to sharply characterize the limits of a broad class of convex relaxations. Our result shows statistical optimality needs to be compromised for achieving computational tractability using convex relaxations. Compared with existing results on computational lower bounds for statistical problems, which consider general polynomial-time algorithms and rely on computational hardness hypotheses on problems like planted clique detection, our theory focuses on a broad class of convex relaxations and does not rely on unproven hypotheses.

We consider two closely related problems: planted clustering and submatrix localization. The planted clustering problem assumes that a random graph is generated based on some underlying clusters of the nodes; the task is to recover these clusters given the graph. The submatrix localization problem concerns locating hidden submatrices with elevated means inside a large real-valued random matrix. Of particular interest is the setting where the number of clusters/submatrices is allowed to grow unbounded with the problem size. These formulations cover several classical models such as planted clique, planted densest subgraph, planted partition, planted coloring, and stochastic block model, which are widely used for studying community detection and clustering/bi-clustering. For both problems, we show that the space of the model parameters (cluster/submatrix size, cluster density, and submatrix mean) can be partitioned into four disjoint regions corresponding to decreasing statistical and computational complexities: (1) the \emph{impossible} regime, where all algorithms fail; (2) the \emph{hard} regime, where the computationally expensive Maximum Likelihood Estimator (MLE) succeeds; (3) the \emph{easy} regime, where the polynomial-time convexified MLE succeeds; (4) the \emph{simple} regime, where a simple counting/thresholding procedure succeeds. Moreover, we show that each of these algorithms provably fails in the previous harder regimes. Our theorems establish the minimax recovery limit, which are tight up to constants and hold with a growing number of clusters/submatrices, and provide a stronger performance guarantee than previously known for polynomial-time algorithms. Our study demonstrates the tradeoffs between statistical and computational considerations, and suggests that the minimax recovery limit may not be achievable by polynomial-time algorithms.

We obtain the conditions for the emergence of the swarm intelligence effect in an interactive game of restless multi-armed bandit (rMAB). A player competes with multiple agents. Each bandit has a payoff that changes with a probability $p_{c}$ per round. The agents and player choose one of three options: (1) Exploit (a good bandit), (2) Innovate (asocial learning for a good bandit among $n_{I}$ randomly chosen bandits), and (3) Observe (social learning for a good bandit). Each agent has two parameters $(c,p_{obs})$ to specify the decision: (i) $c$, the threshold value for Exploit, and (ii) $p_{obs}$, the probability for Observe in learning. The parameters $(c,p_{obs})$ are uniformly distributed. We determine the optimal strategies for the player using complete knowledge about the rMAB. We show whether or not social or asocial learning is more optimal in the $(p_{c},n_{I})$ space and define the swarm intelligence effect. We conduct a laboratory experiment (67 subjects) and observe the swarm intelligence effect only if $(p_{c},n_{I})$ are chosen so that social learning is far more optimal than asocial learning.

Kernel approximation via nonlinear random feature maps is widely used in speeding up kernel machines. There are two main challenges for the conventional kernel approximation methods. First, before performing kernel approximation, a good kernel has to be chosen. Picking a good kernel is a very challenging problem in itself. Second, high-dimensional maps are often required in order to achieve good performance. This leads to high computational cost in both generating the nonlinear maps, and in the subsequent learning and prediction process. In this work, we propose to optimize the nonlinear maps directly with respect to the classification objective in a data-dependent fashion. The proposed approach achieves kernel approximation and kernel learning in a joint framework. This leads to much more compact maps without hurting the performance. As a by-product, the same framework can also be used to achieve more compact kernel maps to approximate a known kernel. We also introduce Circulant Nonlinear Maps, which uses a circulant-structured projection matrix to speed up the nonlinear maps for high-dimensional data.