Ashish Sabharwal

Towards understanding and harnessing the potential of clause learning

Effcient implementations of DPLL with the addition of clause learning are the fastest complete Boolean satisfiability solvers and can handle many significant real-world problems, such as verification, planning and design. Despite its importance, little is known of the ultimate strengths and limitations of the technique. This paper presents the first precise characterization of clause learning as a proof system (CL), and begins the task of understanding its power by relating it to the well-studied resolution proof system. In particular, we show that with a new learning scheme, CL can provide exponentially shorter proofs than many proper refinements of general resolution (RES) satisfying a natural property. These include regular and Davis-Putnam resolution, which are already known to be much stronger than ordinary DPLL. We also show that a slight variant of CL with unlimited restarts is as powerful as RES itself. Translating these analytical results to practice, however, presents a challenge because of the nondeterministic nature of clause learning algorithms. We propose a novel way of exploiting the underlying problem structure, in the form of a high level problem description such as a graph or PDDL specification, to guide clause learning algorithms toward faster solutions. We show that this leads to exponential speed-ups on grid and randomized pebbling problems, as well as substantial improvements on certain ordering formulas.

Algorithm selection and scheduling

Algorithm portfolios aim to increase the robustness of our ability to solve problems efficiently. While recently proposed algorithm selection methods come ever closer to identifying the most appropriate solver given an input instance, they are bound to make wrong and, at times, costly decisions. Solver scheduling has been proposed to boost the performance of algorithm selection. Scheduling tries to allocate time slots to the given solvers in a portfolio so as to maximize, say, the number of solved instances within a given time limit. We show how to solve the corresponding optimization problem at a low computational cost using column generation, resulting in fast and high quality solutions. We integrate this approach with a recently introduced algorithm selector, which we also extend using other techniques. We propose various static as well as dynamic scheduling strategies, and demonstrate that in comparison to pure algorithm selection, our novel combination of scheduling and solver selection can significantly boost performance.

Chapter 2 Satisfiability Solvers

Publisher Summary The past few years have seen enormous progress in the performance of Boolean satisfiability (SAT) solvers. Despite the worst-case exponential run time of all known algorithms, satisfiability solvers are increasingly leaving their mark as a general-purpose tool in areas as diverse as software and hardware verification, automatic test-pattern generation, planning, scheduling, and even challenging problems from algebra. Annual SAT competitions have led to the development of dozens of clever implementations of such solvers, exploration of new techniques, and creation of an extensive suite of real-world instances as well as challenging hand-crafted benchmark problems. Modern SAT solvers provide a black-box procedure that can often solve hard structured problems with over a million variables and several million constraints. This chapter describes the main solution techniques used in modern SAT solvers, classifying them as complete and incomplete methods. It discusses recent insights explaining the effectiveness of these techniques on practical SAT encodings and presents several extensions of the SAT approach currently under development. These extensions further expand the range of applications to include multiagent and probabilistic reasoning.

Using Problem Structure for Efficient Clause Learning

DPLL based clause learning algorithms for satisfiability testing are known to work very well in practice. However, like most branch-and-bound techniques, their performance depends heavily on the variable order used in making branching decisions. We propose a novel way of exploiting the underlying problem structure to guide clause learning algorithms toward faster solutions. The key idea is to use a higher level problem description, such as a graph or a PDDL specification, to generate a good branching sequence as an aid to SAT solvers. The sequence captures hierarchical structure that is lost in the CNF translation. We show that this leads to exponential speedups on grid and randomized pebbling problems. The ideas we use originate from the analysis of problem structure recently used in [1] to study clause learning from a theoretical perspective.

Parallel SAT solver selection and scheduling

Combining differing solution approaches by means of solver portfolios has proven as a highly effective technique for boosting solver performance. We consider the problem of generating parallel SAT solver portfolios. Our approach is based on a recently introduced sequential SAT solver portfolio that excelled at the last SAT competition. We show how the approach can be generalized for the parallel case, and how obstacles like parallel SAT solvers and symmetries induced by identical processors can be overcome. We compare different ways of computing parallel solver portfolios with the best performing parallel SAT approaches to date. Extensive experimental results show that the developed methodology very significantly improves our current parallel SAT solving capabilities.

Guiding combinatorial optimization with UCT

We propose a new approach for search tree exploration in the context of combinatorial optimization, specifically Mixed Integer Programming (MIP), that is based on UCT, an algorithm for the multi-armed bandit problem designed for balancing exploration and exploitation in an online fashion. UCT has recently been highly successful in game tree search. We discuss the differences that arise when UCT is applied to search trees as opposed to bandits or game trees, and provide initial results demonstrating that the performance of even a highly optimized state-of-the-art MIP solver such as CPLEX can be boosted using UCTs guidance on a range of problem instances.

Message-passing and local heuristics as decimation strategies for satisfiability

Decimation is a simple process for solving constraint satisfaction problems, by repeatedly fixing variable values and simplifying without reconsidering earlier decisions. We investigate different decimation strategies, contrasting those based on local, syntactic information from those based on message passing, such as statistical physics based Survey Propagation (SP) and the related and more well-known Belief Propagation (BP). Our results reveal that once we resolve convergence issues, BP itself can solve fairly hard random k-SAT formulas through decimation; the gap between BP and SP narrows down quickly as k increases. We also investigate observable differences between BP/SP and other common CSP heuristics as decimation proceeds, exploring the hardness of the decimated formulas and identifying a somewhat unexpected feature of message passing heuristics, namely, unlike other heuristics for satisfiability, they avoid unit propagation as variables are fixed.

Non-model-based algorithm portfolios for SAT

When tackling a computationally challenging combinatorial problem, one often observes that some solution approaches work well on some instances, while other approaches work better on other instances. This observation has given rise to the idea of building algorithm portfolios [5]. Leyton-Brown et al. [1], for instance, proposed to select one of the algorithms in the portfolio based on some features of the instance to be solved. This approach has been blessed with tremendous success in the past. Especially in SAT, the SATzilla portfolios [7] have performed extremely well in past SAT Competitions [6].

Constraint reasoning and Kernel clustering for pattern decomposition with scaling

Motivated by an important and challenging task encountered in material discovery, we consider the problem of finding K basis patterns of numbers that jointly compose N observed patterns while enforcing additional spatial and scaling constraints. We propose a Constraint Programming (CP) model which captures the exact problem structure yet fails to scale in the presence of noisy data about the patterns. We alleviate this issue by employing Machine Learning (ML) techniques, namely kernel methods and clustering, to decompose the problem into smaller ones based on a global data-driven view, and then stitch the partial solutions together using a global CP model. Combining the complementary strengths of CP and ML techniques yields a more accurate and scalable method than the few found in the literature for this complex problem.

Connections in networks: a hybrid approach

This paper extends our previous work by exploring the use of a hybrid solution method for solving the connection subgraph problem. We employ a two phase solution method, which drastically reduces the cost of testing for infeasibility and also helps prune the search space for MIP-based optimization. Overall, this provides a much more scalable solution than simply optimizing a MIP model of the problem with Cplex. We report results for semi-structured lattice instances as well as on real data used for the construction of a wildlife corridor for grizzly bears in the Northern Rockies region.