Using Learning Automata to Enhance Local-Search Based SAT Solvers with Learning Capability
نویسندگان
چکیده
The conflict between exploration and exploitation is a well-known problem in machine learning and other areas of artificial intelligence. Learning Automata (LA) Thathachar & Sastry (2004); Tsetlin (1973) capture the essence of this conflict, and have thus occupied researchers for over forty years. Initially, LA were used to model biological systems, however, in the last decades they have also attracted considerable interest because they can learn the optimal action when operating in unknown stochastic environments. Furthermore, they combine rapid and accurate convergence with low computational complexity. Recent applications of LA include allocation of polling resources in web monitoring Granmo et al. (2007), allocation of limited sampling resources in binomial estimation Granmo et al. (2007), and optimization of throughput in MPLS traffic engineering Oommen et al. (2007). LA solutions have furthermore found application within combinatorial optimization problems. In Gale et al. (1990); Oommen & Ma (1988) a so-called Object Migration Automaton is used for solving the classical equipartitioning problem. An order of magnitude faster convergence is reported compared to the best known algorithms at that time. A similar approach has also been discovered for the Graph Partitioning ProblemOommen & Croix (1996). Finally, the list organization problem has successfully been addressed by LA schemes. These schemes have been found to converge to the optimal arrangement with probability arbitrary close to unity Oommen & Hansen (1987). In this chapter we study a new application domain for LA, namely, the Satisfiability (SAT) problem. In brief, we demonstrate how the learning capability of LA can be incorporated into selected classical local-search based solvers for SAT-like problems, with the purpose of allowing improved performance by means of learning. Furthermore, we provide a detailed empirical evaluation of the resulting LA based algorithms, and we analyze the effect that the introduced learning has on the local-search based solvers.
منابع مشابه
Structure-Based Satisfiability Checking
Constraint satisfaction deals with developing automated techniques for solving computationally hard problems in a declarative fashion. This note summarizes a dissertation that focuses on search-based methods for the propositional satisfiability problem (SAT). As methods for propositional satisfiability checking have rapidly progressed during the last 15 years, implementations of decision proced...
متن کاملA Decision-Making Procedure for Resolution-Based SAT-Solvers
We describe a new decision-making procedure for resolutionbased SAT-solvers called Decision Making with a Reference Point (DMRP). In DMRP, a complete assignment called a reference point is maintained. DMRP is aimed at finding a change of the reference point under which the number of clauses falsified by the modified point is smaller than for the original one. DMRP makes it possible for a DPLL-l...
متن کاملThe long way from CDClL
Current SAT solvers are powerful enough to be used as engines in real applications. Those applications made the success of a special kind of SAT solvers, namely Conflict Driven Clause Learning SAT solvers (CDClL for short), developed initially by Joao Marques Silva with GRASP [8], and popularized by the SAT solver Chaff [9]. Despite SAT being a NP-complete problem in theory, it might look tract...
متن کاملReinforcement Learning and Reactive Search: an adaptive MAX-SAT solver
This paper investigates Reinforcement Learning (RL) applied to online parameter tuning in Stochastic Local Search (SLS) methods. In particular, a novel application of RL is proposed in the Reactive Tabu Search (RTS) scheme, where the appropriate amount of diversification in prohibition-based local search is adapted in a fast online manner to the characteristics of a task and of the local config...
متن کاملOff the Trail: Re-examining the CDCL Algorithm
Most state of the art SAT solvers for industrial problems are based on the Conflict Driven Clause Learning (CDCL) paradigm. Although this paradigm evolved from the systematic DPLL search algorithm, modern techniques of far backtracking and restarts make CDCL solvers non-systematic. CDCL solvers do not systematically examine all possible truth assignments as does DPLL. Local search solvers are a...
متن کامل