Abstract-The longest common subsequence problem is classical string problem. It has applications, for example, in pattern recognition and bioinformatics. In this work we present a socalled Beam-ACO approach for solving this problem. Beam-ACO algorithms are hybrid techniques that results from a combination of ant colony optimization and beam search, which is an incomplete branch and bound derivative. Our results show that Beam-ACO is able to find new best solutions for 31 out of 60 benchmark instances that we chose for the experimental evaluation of the algorithm.
I. INTRODUCTIONThe longest common subsequence (LCS) problem is one of the classical string problems. Given a problem instance (S, Σ), where S = {s 1 , s 2 , . . . , s n } is a set of n strings over a finite alphabet Σ, the problem consists in finding a longest string t * that is a subsequence of all the strings in S. Such a string t * is called a longest common subsequence of the strings in S. Note that a string t is called a subsequence of a string s, if t can be produced from s by deleting characters. For example, dga is a subsequence of adagtta. Due to its classical nature, the LCS problem has attracted quite a lot of research efforts over the past decades. Much work has been dedicated to the development of efficient dynamic programming procedures (see, for example, [2]). The body of work on approximate methods is dominated by constructive one-pass heuristics [9], [10]. Moreover, metaheuristics have been proposed in [17], [8], [12]. In [3] we recently published the current state-of-the-art algorithm for the LCS problem. The results of this algorithm, which is based on beam search (BS), have shown that none of the earlier algorithms came even close to derive good solutions for difficult problem instances. In other words, BS has shown to be largely superior to any other existing technique for what concerns the LCS problem. In this work we try to improve over the state-of-the-art results of BS by adding a learning component to BS. The resulting algorithm is a hybrid between the metaheuristic ant colony optimization (ACO) [7] and beam search. The interested reader might note that we already published a preliminary version of this