Discussion of "Latent variable graphical model selection via convex optimization" by Venkat Chandrasekaran, Pablo A. Parrilo and Alan S. Willsky [arXiv:1008.1290].

Discussion of "Latent variable graphical model selection via convex optimization" by Venkat Chandrasekaran, Pablo A. Parrilo and Alan S. Willsky [arXiv:1008.1290].

We study the mixtures of factorizing probability distributions represented as visible marginal distributions in stochastic layered networks. We take the perspective of kernel transitions of distributions, which gives a unified picture of distributed representations arising from Deep Belief Networks (DBN) and other networks without lateral connections. We describe combinatorial and geometric properties of the set of kernels and products of kernels realizable by DBNs as the network parameters vary. We describe explicit classes of probability distributions, including exponential families, that can be learned by DBNs. We use these submodels to bound the maximal and the expected Kullback-Leibler approximation errors of DBNs from above depending on the number of hidden layers and units that they contain.

Discussion of "Latent variable graphical model selection via convex optimization" by Venkat Chandrasekaran, Pablo A. Parrilo and Alan S. Willsky [arXiv:1008.1290].

Learning new motor tasks autonomously from interaction with a human being is an important goal for both robotics and machine learning. However, when moving beyond basic skills, most monolithic machine learning approaches fail to scale. In this paper, we take the task of learning table tennis as an example and present a new framework which allows a robot to learn cooperative table tennis from interaction with a human. Therefore, the robot first learns a set of elementary table tennis hitting movements from a human teacher by kinesthetic teach-in, which is compiled into a set of dynamical system motor primitives (DMPs). Subsequently, the system generalizes these movements to a wider range of situations using our mixture of motor primitives (MoMP) approach. The resulting policy enables the robot to select appropriate motor primitives as well as to generalize between them. Finally, the robot plays with a human table tennis partner and learns online to improve its behavior.

We present an approach for learning grounded language from mixed-initiative human-robot interaction. Prior work on learning from human instruction has concentrated on acquisition of task-execution knowledge from domain-specific language. In this work, we demonstrate acquisition of linguistic, semantic, perceptual, and procedural knowledge from mixed-initiative, natural language dialog. Our approach has been instantiated in a cognitive architecture, Soar, and has been deployed on a table-top robotic arm capable of picking up small objects. A preliminary analysis verifies the ability of the robot to acquire diverse knowledge from human-robot interaction.

A large number of transients visit big cities on any given day and they visit crowded areas and come in contact with many people. However, epidemiological studies have not paid much attention to the role of this subpopulation in disease spread. In the present work, we extend a synthetic population model of Washington DC metro area to include leisure and business travelers. This approach involves combining Census data, activity surveys, and geospatial data to build a detailed minute-by-minute simulation of population interaction. We simulate a flu-like disease outbreak both with and without the transient population to evaluate the effect of the transients on outbreak size and peak day in terms of number of residents infected. Results show that there are significantly more infections when transients are considered. We also evaluate interventions like closing big museums and encouraging use of hand sanitizers at those musuems. Surprisingly closing musuems does not result in a significant difference in the epidemic. However, we find that if the use of hand sanitizer reduces the infectivity and suceptibility to 80% or 60% of the original values, it is as effective as closing museums for a few days or entirely eliminating the effect of transients. If infectivity and susceptibility are reduced to 40% or 20%, it reduces the number of resident infections over the period of 120 days by 10% and 13%.

In this paper, we present an optimized version of the previously developed Block Modularity algorithm (Anthony,2009). The original algorithm was a fast, greedy method that effectively discovered a structured clustering in linked data and scaled very well with the number of nodes and edges. The optimized version is scalable in terms of the model complexity; the technique can now be used effectively to discover thousands of clusters in data sets with hundreds of thousands (and possibly more) nodes and edges. The optimization leads to an improvement of the runtime per iteration from cubic to quadratic with a small increase in the constant factor. The algorithm compares favorably with Karrer and Newman's Degree-Corrected Block Model (DCBM) in both runtime and quality of results.

Human control of vehicle swarms faces a dilemma: an operator must be able to exercise precise control over how a mission is executed, but controlling individual vehicles is not scalable. The Proto spatial computing lan- guage offers an intermediate representation, where the motion of a swarm is specified as a vector field, which is then approximated by the movement of individual members (Bachrach, Beal, and McLurkin 2010). I propose that this can be exploited to build a “tactical command” model of swarm control, whereby human “officers” dynamically decompose a swarm into units and task those units to carry out geometric and topological maneuvers under the constraints imposed by the platform. This abstraction may also allow situation awareness interfaces for individual agents to be extended to apply to swarm units.

Current efforts aim to incorporate knowledge from clinical practice guidelines (CPGs) into computer systems using sophisticated interchange formats. Due to their complexity, such formats require expensive manual formalization work. This paper presents a preliminary study of using natural language processing (NLP) to automatically formalize CPG recommendations. We developed a CPG representation using concepts from the Systematized Nomenclature of Medicine – Clinical Terms (SNOMED–CT), and manually applied this representation to a sample of CPG recommendations that is representative of multiple medical domains and recommendation types. Using this resource, we trained and evaluated a supervised classification model that formalizes new CPG recommendations according to the SNOMED–CT representation, achieving a precision of 75% and recall of 42% (F1 = 54%). We have identified two important lines of future investigation: (1) feature engineering to address the unique linguistic properties of CPG recommendations, and (2) alternative model formulations that are more robust to processing errors. A third line of investigation – creating additional training data for the NLP model – is shown to be of little utility.