In this work we present an advanced Bayesian formulation to the task of control learning that employs the Relevance Vector Machines (RVM) generative model for value function evaluation. The key aspect of the proposed method is the design of the discount return as a generalized linear model that constitutes a well-known probabilistic approach. This allows to augment the model with advantageous sparse priors provided by the RVM's regression framework. We have also taken into account the significant issue of selecting the proper parameters of the kernel design matrix. Experiments have shown that our method produces improved performance in both simulated and real test environments.

One proposed approach to managing a large complex Smart Grid is through Broker Agents who buy electrical power from distributed producers, and also sell power to consumers, via a Tariff Market--a new market mechanism where Broker Agents publish concurrent bid and ask prices. A key challenge is the specification of the market strategy that the Broker Agents should use in order to earn profits while maintaining the market's balance of supply and demand. Interestingly, previous work has shown that a Broker Agent can learn its strategy, using Markov Decision Processes (MDPs) and Q-learning, and outperform other Broker Agents that use predetermined or randomized strategies. In this work, we investigate the more representative scenario in which multiple Broker Agents, instead of a single one, are independently learning their strategies. Using a simulation environment based on real data, we find that Broker Agents who employ periodic increases in exploration achieve higher rewards. We also find that varying levels of market dominance in customer allocation models result in remarkably distinct outcomes in market prices and aggregate Broker Agent rewards. The latter set of results can be explained by established economic principles regarding the emergence of monopolies in market-based competition, further validating our approach.

When mobile robots perform tasks in environments with humans, it seems appropriate for the robots to rely on such humans for help instead of dedicated human oracles or supervisors. However, these humans are not always available nor always accurate. In this work, we consider human help to a robot as concretely providing observations about the robot's state to reduce state uncertainty as it executes its policy autonomously. We model the probability of receiving an observation from a human in terms of their availability and accuracy by introducing Human Observation Providers POMDPs (HOP-POMDPs). We contribute an algorithm to learn human availability and accuracy online while the robot is executing its current task policy. We demonstrate that our algorithmis effective in approximating the true availability and accuracy of humans without depending on oracles to learn, thus increasing the tractability of deploying a robot that can occasionally ask for help.

This paper extends existing plan recognition research to handle plans containing loops. We supply an encoding of plans with loops for recognition, based on techniques used to parse lexicalized grammars, and demonstrate its effectiveness empirically. To do this, the paper first shows how encoding plan libraries as context free grammars permits the application of standard rewriting techniques to remove left recursion and ε-productions, thereby enabling polynomial time parsing. However, these techniques alone fail to provide efficient algorithms for plan recognition. We show how the loop-handling methods from formal grammars can be extended to the more general plan recognition problem and provide a method for encoding loops in an existing plan recognition system that scales linearly in the number of loop iterations.

We present a first theoretical analysis of the power of polynomial-time preprocessing for important combinatorial problems from various areas in AI. We consider problems from Constraint Satisfaction, Global Constraints, Satisfiability, Nonmonotonic and Bayesian Reasoning. We show that, subject to a complexity theoretic assumption, none of the considered problems can be reduced by polynomial-time preprocessing to a problem kernel whose size is polynomial in a structural problem parameter of the input, such as induced width or backdoor size. Our results provide a firm theoretical boundary for the performance of polynomial-time preprocessing algorithms for the considered problems.

In many multi-agent applications such as distributed sensor nets, a network of agents act collaboratively under uncertainty and local interactions. Networked Distributed POMDP (ND-POMDP) provides a framework to model such cooperative multi-agent decision making. Existing work on ND-POMDPs has focused on offline techniques that require accurate models, which are usually costly to obtain in practice. This paper presents a model-free, scalable learning approach that synthesizes multi-agent reinforcement learning (MARL) and distributed constraint optimization (DCOP). By exploiting structured interaction in ND-POMDPs, our approach distributes the learning of the joint policy and employs DCOP techniques to coordinate distributed learning to ensure the global learning performance. Our approach can learn a globally optimal policy for ND-POMDPs with a property called groupwise observability. Experimental results show that, with communication during learning and execution, our approach significantly outperforms the nearly-optimal non-communication policies computed offline.

Many algorithms for performing inference in graphical models have complexity that is exponential in the treewidth — a parameter of the underlying graph structure. Computing the (minimal) treewidth is NPcomplete, so stochastic algorithms are sometimes used to find low width tree decompositions. A common approach for finding good decompositions is iteratively executing a greedy triangulation algorithm (e.g. minfill) with randomized tie-breaking. However, utilizing a stochastic algorithm as part of the inference task introduces a new problem — namely, deciding how long the stochastic algorithm should be allowed to execute before performing inference on the best tree decomposition found so far. We refer to this dilemma as the Stopping Problem and formalize it in terms of the total time needed to answer a probabilistic query. We propose a rule for discontinuing the search for improved decompositions and demonstrate the benefit (in terms of time saved) of applying this rule to Bayes and Markov network instances.

The behavior and dynamics of complex systems are in focus of many research fields. The complexity of such systems comes not only from the number of their elements, but also from the unavoidable emergence of new properties of the system, which are not just a simple summation of the properties of its elements. The behavior of complex systems can be fitted with a number of well developed models, which, however, do not incorporate the modularity and the evolution of a system simultaneously. In this work, we propose a generalized model that addresses this issue. Our model is developed within the Random Set Theory’s framework and allows for reconstructing the stochastic evolution diagrams of complex systems.

Utility elicitation is an important component of many applications, such as decision support systems and recommender systems. Such systems query users about their preferences and offer recommendations based on the system's belief about the user's utility function. We analyze the connection between the problem of generating optimal recommendation sets and the problem of generating optimal choice queries, considering both Bayesian and regret-based elicitation. Our results show that, somewhat surprisingly, under very general circumstances, the optimal recommendation set coincides with the optimal query.

Recently, a number of researchers have proposed spectral algorithms for learning models of dynamical systems — for example, Hidden Markov Models (HMMs), Partially Observable Markov Decision Processes (POMDPs), and Transformed Predictive State Representations (TPSRs). These algorithms are attractive since they are statistically consistent and not subject to local optima. However, they are batch methods: they need to store their entire training data set in memory at once and operate on it as a large matrix, and so they cannot scale to extremely large data sets (either many examples or many features per example). In turn, this restriction limits their ability to learn accurate models of complex systems. To overcome these limitations, we propose a new online spectral algorithm, which uses tricks such as incremental Singular Value Decomposition (SVD) and random projections to scale to much larger data sets and more complex systems than previous methods. We demonstrate the new method on an inertial measurement prediction task and a high-bandwidth video mapping task and we illustrate desirable behaviors such as "closing the loop," where the latent state representation changes suddenly as the learner recognizes that it has returned to a previously known place.