We report on Jim, an inexpensive student designed platform for embodied affective AI. The project brings together students from backgrounds in computer science, physics, engineering, and Digital Media Arts (DMA) in an informal educational setting. The platform will be used in AI courses and autism treatment studies.

Pattern discovery in protein interaction networks can reveal crucial biological knowledge on the inner workings of cellular machinery. Although far from complete, extracting meaningful patterns from proteomic networks is a non-trivial task due to their size-complexity. This paper proposes a computational framework to efficiently discover topologically-similar patterns from large proteomic networks using Particle Swarm Optimization (PSO). PSO is a robust and low-cost optimization technique that demonstrated to work effectively on the complex, mostly sparse proteomic networks. The resulting topologically-similar patterns of close proximity are utilized to systematically predict new high-confidence protein-protein interactions (PPIs). The proposed PSO-based PPI prediction method (3PI) managed to predict high-confidence PPIs, validated by more than one computational/experimental source, through a proposed PPI knowledge transfer process between topologically-similar interaction patterns of close proximity. In three case studies, over 50% of the predicted interactions for EFGR, ERBB2, ERBB3, GRB2 and UBC are overlapped with publically available interaction databases, ~80% of the predictions are found among the Top 1% results of another PPI prediction method and their genes are significantly co-expressed across different tissues. Moreover, the only single prediction example that did not overlap with any of our validation sources was recently experimentally supported by two PubMed publications.

We analyze and evaluate a generative process for multiple-instance learning (MIL) in which bags are distributions over instances. We show that our generative process contains as special cases generative models explored in prior work, while excluding scenarios known to be hard for MIL. Further, under the mild assumption that every negative instance is observed with nonzero probability in some negative bag, we show that it is possible to learn concepts that accurately label instances from MI data in this setting. Finally, we show that standard supervised approaches can learn concepts with low area-under-ROC error from MI data in this setting. We validate this surprising result with experiments using several synthetic and real-world MI datasets that have been annotated with instance labels.

Many multi-agent coordination problems can be represented as DCOPs. Motivated by task allocation in disaster response, we extend standard DCOP models to consider uncertain task rewards where the outcome of completing a task depends on its current state, which is randomly drawn from unknown distributions. The goal of solving this problem is to find a solution for all agents that minimizes the overall worst-case loss. This is a challenging problem for centralized algorithms because the search space grows exponentially with the number of agents and is nontrivial for existing algorithms for standard DCOPs. To address this, we propose a novel decentralized algorithm that incorporates Max-Sum with iterative constraint generation to solve the problem by passing messages among agents. By so doing, our approach scales well and can solve instances of the task allocation problem with hundreds of agents and tasks.

Mobile-manipulation robots performing service tasks in human-centric indoor environments has long been a dream for developers of autonomous agents. Tasks such as cooking and cleaning require interaction with the environment, hence robots need to know relevant aspects of their spatial surroundings. However, unlike the structured settings that industrial robots operate in, service robots typically have little prior information about their environment. Even if this information was given, due to the involvement of many other agents (e.g., humans moving objects), uncertainty in the complete state of the world is inevitable over time. Additionally, most information about the world is irrelevant to any particular task at hand. Mobile manipulation robots therefore need to continuously perform the task of state estimation, using perceptual information to maintain the state, and its uncertainty, of task-relevant aspects of the world. Because indoor tasks frequently require the use of objects, objects should be given critical emphasis in spatial representations for service robots. Compared to occupancy grids and feature-based maps often used in navigation and SLAM, object-based representations are arguably still in their infancy. In my thesis, I propose a representation framework based on objects, their 'semantic' attributes, and their geometric realizations in the physical world.

Conformant probabilistic planning (CPP) differs from conformant planning (CP) by two key elements: the initial belief state is probabilistic,and the conformant plan must achieve the goal with probability $\geq\theta$, for some $0<\theta\leq 1$. In earlier work we observed that one can reduce CPP to CP by finding a set of initial states whose probability $\geq\theta$, for whicha conformant plan exists. In previous solvers we used the underlying planner to select this set of states and to plan for them simultaneously. Here we suggest an alternative approach: start with relevance analysis to determine a promising set of initial states on which to focus. Then, call an off-the-shelf conformant planner to solve the resulting problem. This approach has a number of advantages. First, instead of depending on the heuristic function to select the set of initial states,we can introduce specific, efficient relevance reasoning techniques. Second, we can benefit from optimizations used by conformant planners that are unsound when applied to the original CPP. Finally, we are free to use any existing (or new) CP solver. Consequently, the new planner dominates previous solvers on almost all domains and scales to instances that were not solved before.

Decomposition is a technique to obtain complete solutions by assembling independently obtained partial solutions. In particular, constraint decomposition plays an important role in distributed databases, distributed scheduling and violation detection: It enables conflict-free local decision making, while avoiding communication overloading. One of the main issues in decomposition is the loss of flexibility due to decomposition. Here, flexibility roughly refers to the freedom in choosing suitable values for the variables in order to satisfy the constraints. In this paper, we concentrate on linear constraint systems and efficient decomposition techniques for them. Using a generalization of a flexibility metric developed for Simple Temporal Networks, we show how an efficient decomposition technique for linear constraint systems can be derived that minimizes the loss of flexibility. As a by-product of this decomposition technique, we propose an intuitively attractive flexibility metric for linear constraint systems where decomposition does not incur any loss of flexibility.

The automatic discovery of a significant low-dimensional feature representation from a given data set is a fundamental problem in machine learning. This paper focuses specifically on the development of the feature representation discovery methods appropriate for high-dimensional and sparse data. We formulate our feature representation discovery problem as a variant of the semi-supervised learning problem, namely, as an optimization problem over unsupervised data whose objective is evaluating the impact of each feature with respect to modeling a target task according to the initial model constructed by using supervised data. The most notable characteristic of our method is that it offers a feasible processing speed even if the numbers of data and features are both in the millions or even billions, and successfully provides a significantly small number of feature sets, i.e., fewer than 10, that can also offer improved performance compared with those obtained with the original feature sets. We demonstrate the effectiveness of our method in experiments consisting of two well-studied natural language processing tasks.

Most active learning methods avoid model selection by training models of one type (SVMs, boosted trees, etc.) using one pre-defined set of model hyperparameters. We propose an algorithm that actively samples data to simultaneously train a set of candidate models (different model types and/or different hyperparameters) and also select the best model from this set. The algorithm actively samples points for training that are most likely to improve the accuracy of the more promising candidate models, and also samples points for model selection---all samples count against the same labeling budget. This exposes a natural trade-off between the focused active sampling that is most effective for training models, and the unbiased sampling that is better for model selection. We empirically demonstrate on six test problems that this algorithm is nearly as effective as an active learning oracle that knows the optimal model in advance.

In heterogeneous multi-robot teams, robustness and flexibility are increased by the diversity of the robots, each contributing different capabilities. Yet platform-independence is desirable when planning actions for the various robots. We propose a platform-independent model of robot capabilities which we use as a planning domain. We extend existing planning techniques to support two requirements: generating new objects during planning; and, required concurrency of actions due to data flow which can be cyclic. The first requires online action instantiation, the second a small extension of the Planning Domain Definition Language (PDDL): allowing predicates in continuous effects. We evaluate the planner on benchmark domains and present results on an example object transportation task in simulation.