Because an agent's resources dictate what actions it can possibly take, it should plan which resources it holds over time carefully, considering its inherent limitations (such as power or payload restrictions), the competing needs of other agents for the same resources, and the stochastic nature of the environment. Such agents can, in general, achieve more of their objectives if they can use -- and even create -- opportunities to change which resources they hold at various times. Driven by resource constraints, the agents could break their overall missions into an optimal series of phases, optimally reconfiguring their resources at each phase, and optimally using their assigned resources in each phase, given their knowledge of the stochastic environment. In this paper, we formally define and analyze this constrained, sequential optimization problem in both the single-agent and multi-agent contexts. We present a family of mixed integer linear programming (MILP) formulations of this problem that can optimally create phases(when phases are not predefined) accounting for costs and limitations in phase creation. Because our formulations simultaneously also find the optimal allocations of resources at each phase and the optimal policies for using the allocated resources at each phase, they exploit structure across these coupled problems. This allows them to find solutions significantly faster (orders of magnitude faster in larger problems) than alternative solution techniques, as we demonstrate empirically.

Fully characterizing structural and functional sites in proteins is a fundamental step in understanding their roles in the cell. This extremely challenging combinatorial problem requires determining the number of sites in the protein and the set of residues involved in each of them. We formulate it as a distance-based supervised clustering task, where training proteins are employed to learn a proper distance function between residues. A partial clustering is then returned by searching for maximum-weight cliques in the resulting weighted graph representation of proteins. A novel stochastic local search algorithm is proposed to efficiently generate approximate solutions. Our method achieves substantial improvements over a previous structured-output approach for metal binding site prediction. Significant improvements over the current state-of-the-art are also achieved in predicting catalytic sites from 3D structure in enzymes.

The feasible solutions of the traveling salesman problem with pickup and delivery (TSPPD) are represented by vertex lists in existing literature. However, when the TSPPD requires that the loading and unloading operations must be performed in a last-in-first-out (LIFO) manner, we show that its feasible solutions can be represented by trees. Consequently, we develop a variable neighbourhood search (VNS) heuristic for the TSPPD with last-in-first-out loading (TSPPDL) involving several search operators based on the tree data structure. Experiments show that our VNS heuristic is superior to the current best heuristics for TSPPDL in terms of both solution quality and computing time.

On the surface, bidirectional search (BDS) is an attractive idea with the potential for significant asymptotic reductions in search effort. However, the results in practice often fall far short of expectations. We introduce a new bidirectional search algorithm, Single-Frontier Bidirectional Searc (SFBDS). Unlike traditional BDS which keeps two frontiers, SFBDS uses a single frontier. Each node in the tree can be seen as an independent task of finding the shortest path between the current start and current goal. At a particular node we can decide to search from start to goal or from goal to start, choosing the direction with the highest potential for minimizing the total work done. Theoretical results give insights as to when this approach will work and experimental data validates the algorithm for a broad range of domains.

The aim of General Game Playing (GGP) is to create intelligent agents that can automatically learn how to play many different games at an expert level without any human intervention. One of the main challenges such agents face is to automatically learn knowledge-based heuristics in real-time, whether for evaluating game positions or for search guidance. In recent years, GGP agents that use Monte-Carlo simulations to reason about their actions have become increasingly more popular. For competitive play such an approach requires an effective search-control mechanism for guiding the simulation playouts. In here we introduce several schemes for automatically learning search guidance based on both statistical and reinforcement learning techniques. We compare the different schemes empirically on a variety of games and show that they improve significantly upon the current state-of-the-art in simulation-control in GGP. For example, in the chess-like game Skirmish, which has proved a particularly challenging game for simulation-based GGP agents, an agent employing one of the proposed schemes achieves 97% winning rate against an unmodified agent.

It is well known that there cannot be a single "best" heuristic for optimal planning in general. One way of overcoming this is by combining admissible heuristics (e.g. by using their maximum), which requires computing numerous heuristic estimates at each state. However, there is a tradeoff between the time spent on computing these heuristic estimates for each state, and the time saved by reducing the number of expanded states. We present a novel method that reduces the cost of combining admissible heuristics for optimal search, while maintaining its benefits. Based on an idealized search space model, we formulate a decision rule for choosing the best heuristic to compute at each state. We then present an active online learning approach for that decision rule, and employ the learned model to decide which heuristic to compute at each state. We evaluate this technique empirically, and show that it substantially outperforms each of the individual heuristics that were used, as well as their regular maximum.

A number of algorithms have been proposed for the Minimum Vertex Cover problem. However, they are far from satisfactory, especially on hard instances. In this paper, we introduce Edge Weighting Local Search (EWLS), a new local search algorithm for the Minimum Vertex Cover problem. EWLS is based on the idea of extending a partial vertex cover into a vertex cover. A key point of EWLS is to find a vertex set that provides a tight upper bound on the size of the minimum vertex cover. To this purpose, EWLS employs an iterated local search procedure, using an edge weighting scheme which updates edge weights when stuck in local optima. Moreover, some sophisticated search strategies have been taken to improve the quality of local optima. Experimental results on the broadly used DIMACS benchmark show that EWLS is competitive with the current best heuristic algorithms, and outperforms them on hard instances. Furthermore, on a suite of difficult benchmarks, EWLS delivers the best results and sets a new record on the largest instance.

People constantly apply acquired knowledge to new learning tasks, but machines almost never do. Research on transfer learning attempts to address this dissimilarity. Working within this area, we report on a procedure that learns and transfers constraints in the context of inductive process modeling, which we review. After discussing the role of constraints in model induction, we describe the learning method, MISC, and introduce our metrics for assessing the cost and benefit of transferred knowledge. The reported results suggest that cross-domain transfer is beneficial in the scenarios that we investigated, lending further evidence that this strategy is a broadly effective means for increasing the efficiency of learning systems. We conclude by discussing the aspects of inductive process modeling that encourage effective transfer, by reviewing related strategies, and by describing future research plans for constraint induction and transfer learning.

The selection of the action to do next is one of the central problems faced by autonomous agents. In AI, three approaches have been used to address this problem: the programming-based approach, where the agent controller is given by the programmer, the learning-based approach, where the controller is induced from experience via a learning algorithm, and the model-based approach, where the controller is derived from a model of the problem. Planning in AI is best conceived as the model-based approach to action selection. The models represent the initial situation, actions, sensors, and goals. The main challenge in planning is computational, as all the models, whether accommodating feedback and uncertainty or not, are intractable in the worst case. In this article, I review some of the models considered in current planning research, the progress achieved in solving these models, and some of the open problems.

The problem of dividing a sequence of values into segments occurs in database systems, information retrieval, and knowledge management. The challenge is to select a finite number of boundaries for the segments so as to optimize an objective error function defined over those segments. Although this optimization problem can be solved in polynomial time, the algorithm which achieves the minimum error does not scale well, hence it is not practical for applications with massive data sets. There is considerable research with numerous approximation and heuristic algorithms. Still, none of those approaches has resolved the quality-efficiency tradeoff in a satisfactory manner. In (Halim, Karras, and Yap 2009), we obtain near linear time algorithms which achieve both the desired scalability and near-optimal quality, thus dominating earlier approaches. In this paper, we show how two ideas from artificial intelligence, an efficient local search and recombination of multiple solutions reminiscent of genetic algorithms, are combined in a novel way to obtain state of the art histogram construction algorithms.