Showing papers by "Satoru Miyano published in 1991"

Teachability in computational learning

Abstract: This paper considers computational learning from the view-point of teaching. We introduce a notion of teachability with which we establish a relationship between the learnability and teachability. We also discuss the complexity issues of a teacher in relation to learning.

A Machine Discovery from Amino Acid Sequences by Decision Trees over Regular Patterns

Abstract: This paper describes a machine learning system that discovered a “negative motif”, in transmembrane domain identification from amino acid sequences, and reports its experiments on protein data using PIR database. We introduce a decision tree whose nodes are labeled with regular patterns. As a hypothesis, the system produces such a decision tree for a small number of randomly chosen positive and negative examples from PIR. Experiments show that our system finds reasonable hypotheses very successfully. As a theoretical foundation, we show that the class of languages defined by decesion trees of depth at mostd overk-variable regular patterns is polynomial-time learnable in the sense of probably approximately correct (PAC) learning for any fixedd, k≥0.

Using Maximal Independent Sets to Solve Problems in Parallel

Abstract: By using an O((log n)2) time EREW PRAM algorithm for a maximal independent set problem (MIS), we show the following two results: (1) Given a graph, the maximal vertex-induced subgraph satisfying a hereditary graph property π can be found in time 0(Δλ(π)Tπ(n)(log n)2) using a polynomial number of processors, where λ(π) is the maximum of diameters of minimal graphs violating π and Tπ(n) is the time needed to decide whether a graph with n vertices satisfies π. (2) Given a family C = c1,...,c m of subsets of a finite set S = 1,..., n with S = (U i=1 m ci, a minimal set cover for S can be computed on an EREW PRAM in time O(αΒ(log(n + m))2) using a polynomial number of processors, where α=max{¦ci,¦ ¦i=1,..., m and Β = max{¦dj¦ ¦j = 1,..., n.

More About Learning Elementary Formal Systems

Abstract: Elementary formal system (EFS for short) is a kind of logic program directly dealing with character strings. In 1989, we proposed the class of variable-bounded EFS's as a unifying framework for language learning. Responding to the proposal, several works have been developed. In this paper, a brief summary of these works on learning elementary formal systems, Shapiro's model inference approach, inductive inference from positive data, Valiant's PAC (probably approximately correct) learning approach, and applications to Molecular Biology, is presented.

A Learning Algorithm for Elementary Formal Systems and its Experiments on Identification of Transmembrane Domains

Abstract: Proposes a method for algorithmic learning of transmembrane domains based on elementary formal systems. An elementary formal system (EFS) is a kind of a logic program consisting of if-then rules. With this framework, the authors have implemented the algorithm for identifying transmembrane domains in amino acid sequences. Because of the limitations on computational resources, they restrict candidate hypotheses to EFSs defined by collections of regular patterns. From 70 transmembrane sequences and a similar amount of negative examples which are not transmembrane sequences, the algorithm has produced several reasonable hypotheses of small size. Experiments with the database PIR show that one of them recognizes 95% of 689 transmembrane sequences and 95% of 19256 negative examples which consist of non-transmembrane sequences of length around 30 randomly chosen from PIR.<>

Analogy is NP-Hard

Abstract: Analogy is described in predicate logic. This paper deals with the analogy without any function symbols except constants. We show that the problem of deciding whether a given atomic formula can be inferred by analogy is NP-hard even in such a simple case.