Actions description languages (ADLs), such as STRIPS, PDDL, and RDDL specify the input format for planning algorithms. Unfortunately, their syntax is familiar to planning experts only, and not to potential users of planning technology. Moreover, this syntax limits the ability to describe complex and large domains. We argue that programming languages (PLs), and more specifically, probabilistic programming languages (PPLs), provide a more suitable alternative. PLs are familiar to all programmers, support complex data types and rich libraries for their manipulation, and have powerful constructs, such as loops, sub-routines, and local variables with which complex, realistic models and complex objectives can be simply and naturally specified. PPLs, specifically, make it easy to specify distributions, which is essential for stochastic models. The natural objection to this proposal is that PLs are opaque and too expressive, making reasoning about them difficult. However, PPLs also come with efficient inference algorithms, which, coupled with a growing body of work on sampling-based and gradient-based planning, imply that planning and execution monitoring can be carried out efficiently in practice. In this paper, we expand on this proposal, illustrating its potential with examples.

We tackle the problem of conditioning probabilistic programs on distributions of observable variables. Probabilistic programs are usually conditioned on samples from the joint data distribution, which we refer to as deterministic conditioning. However, in many real-life scenarios, the observations are given as marginal distributions, summary statistics, or samplers. Conventional probabilistic programming systems lack adequate means for modeling and inference in such scenarios. We propose a generalization of deterministic conditioning to stochastic conditioning, that is, conditioning on the marginal distribution of a variable taking a particular form. To this end, we first define the formal notion of stochastic conditioning and discuss its key properties. We then show how to perform inference in the presence of stochastic conditioning. We demonstrate potential usage of stochastic conditioning on several case studies which involve various kinds of stochastic conditioning and are difficult to solve otherwise. Although we present stochastic conditioning in the context of probabilistic programming, our formalization is general and applicable to other settings.

In this work we show how to represent policies as programs: that is, as stochastic simulators with tunable parameters. To learn the parameters of such policies we develop connections between black box variational inference and existing policy search approaches. We then explain how such learning can be implemented in a probabilistic programming system. Using our own novel implementation of such a system we demonstrate both conciseness of policy representation and automatic policy parameter learning for a set of canonical reinforcement learning problems.

Conflict-Based Search (CBS) and its generalization, Meta-Agent CBS are amongst the strongest newly introduced algorithms for Multi-Agent Path Finding. This paper introduces ICBS, an improved version of CBS. ICBS incorporates three orthogonal improvements to CBS which are systematically described and studied. Experimental results show that each of these improvements reduces the runtime over basic CBS by up to 20x in many cases. When all three improvements are combined, an even larger improvement is achieved, producing state-ofthe art results for a number of domains.

We introduce an approximate search algorithm for fast maximum a posteriori probability estimation in probabilistic programs, which we call Bayesian ascent Monte Carlo (BaMC). Probabilistic programs represent probabilistic models with varying number of mutually dependent finite, countable, and continuous random variables. BaMC is an anytime MAP search algorithm applicable to any combination of random variables and dependencies. We compare BaMC to other MAP estimation algorithms and show that BaMC is faster and more robust on a range of probabilistic models.

Recent advances in metareasoning for search has shown its usefulness in improving numerous search algorithms. This paper applies rational metareasoning to IDA* when several admissible heuristics are available. The obvious basic approach of taking the maximum of the heuristics is improved upon by lazy evaluation of the heuristics, resulting in a variant known as Lazy IDA*. We introduce a rational version of lazy IDA* that decides whether to compute the more expensive heuristics or to bypass it, based on a myopic expected regret estimate. Empirical evaluation in several domains supports the theoretical results, and shows that rational lazy IDA* is a state-of-the-art heuristic combination method.

In this paper we discuss and experiment with Lazy A*, a variant of A* where heuristics are evaluated lazily and with Rational Lazy A*, which decides whether to compute the more expensive heuristics at all, based on a myopic value of information estimate. Full version appears in IJCAI-2013.

UCT, a state-of-the art algorithm for Monte Carlo tree search (MCTS) in games and Markov decision processes, is based on UCB, a sampling policy for the Multi-armed Bandit problem (MAB) that minimizes the cumulative regret. However, search differs from MAB in that in MCTS it is usually only the final ``arm pull'' (the actual move selection) that collects a reward, rather than all ``arm pulls''. Therefore, it makes more sense to minimize the simple regret, as opposed to the cumulative regret. We begin by introducing policies for multi-armed bandits with lower finite-time and asymptotic simple regret than UCB, using it to develop a two-stage scheme (SR+CR) for MCTS which outperforms UCT empirically. Optimizing the sampling process is itself a metareasoning problem, a solution of which can use value of information (VOI) techniques. Although the theory of VOI for search exists, applying it to MCTS is non-trivial, as typical myopic assumptions fail. Lacking a complete working VOI theory for MCTS, we nevertheless propose a sampling scheme that is ``aware'' of VOI, achieving an algorithm that in empirical evaluation outperforms both UCT and the other proposed algorithms.

UCT, a state-of-the art algorithm for Monte Carlo tree search (MCTS),is based on UCB, a policy for the Multi-armed Bandit problem (MAB) thatminimizes the cumulative regret. However, search differs from MAB inthat in MCTS it is usually only the final ``arm pull''that collects a reward, rather than all ``arm pulls''.Therefore, it makes more sense to minimize the simple, rather thancumulative, regret. We introduce policies formulti-armed bandits with lower simpleregret than UCB and develop a two-stage scheme (SR+CR) for MCTSwhich outperforms UCT empirically. We also propose a samplingscheme based on value of information (VOI), achieving an algorithmthat empirically outperforms other proposed algorithms.

The following sequential decision problem is considered: given a set of items of unknown utility, an item with as high a utility as possible must be selected ("the selection problem"). Measurements (possibly noisy) of item features prior to selection are allowed at known costs. The goal is to optimize the overall sequential decision process of measurements and selection. Value of information (VOI) is a well-known scheme for selecting measurements, but the intractability of the problem typically leads to using myopic VOI estimates. In the selection problem, myopic VOI frequently badly underestimates the VOI, leading to inferior measurement policies. In this paper, the strict myopic assumption is relaxed into a scheme termed semimyopic, providing a spectrum of methods that can improve the performance of measurement policies. In particular, the efficiently computable method of "blinkered" VOI is proposed, and theoretical bounds for important special cases are examined. Empirical evaluation of "blinkered" VOI in the selection problem with normally distributed item values shows that it performs much better than pure myopic VOI.