Learning Cellular Automata Represented as Logic Programs

2015 Fiscal Year Final Research Report

• Project/Area Number: 26540122
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Intelligent informatics
• Research Institution: National Institute of Informatics
• Principal Investigator: INOUE KATSUMI 国立情報学研究所, 情報学プリンシプル研究系, 教授 (10252321)
• Co-Investigator(Kenkyū-buntansha): SAKAMA CHIAKI 和歌山大学, システム工学部, 教授 (20273873)
• Project Period (FY): 2014-04-01 – 2016-03-31
• Keywords: 機械学習 / セルオートマトン / ブーリアンネットワーク / 状態遷移 / 行動規則学習 / 論理発見 / 遺伝子制御ネットワーク学習

Outline of Final Research Achievements

Learning from Interpretation Transition (LFIT) is a method of unsupervised learning, which learns the dynamics of a system from observed time-series data. LFIT has been developed in three ways: (1) Learning memory-less systems from 1-step state transitions, which contains three different implementations to learn the state transition rules, that is, (a) generalization based on the resolution principle, (b) extension of the binary decision diagram (BDD), and (c) least specialization that guarantees the minimality of learned rules; (2) Learning systems with memory (or delay), which can learn Markov(k) systems that depend on k previous states; and (3) Learning nondeterministic and probabilistic systems, which can work for noisy data.
These three learning algorithms have been implemented and evaluated with bioinformatics data to construct gene regulatory networks. LFIT has also been applied to identification of cellular automata, learning robot planning rules, and learning logics.

Free Research Field

知能情報学