Approach from learning theory toward understanding the limitations of computation

2016 Fiscal Year Final Research Report

• Project Area: A multifaceted approach toward understanding the limitations of computation
• Project/Area Number: 24106010
• Research Category: Grant-in-Aid for Scientific Research on Innovative Areas (Research in a proposed research area)
• Allocation Type: Single-year Grants
• Review Section: Science and Engineering
• Research Institution: Kyushu University
• Principal Investigator: Takimoto Eiji 九州大学, システム情報科学研究院, 教授 (50236395)
• Co-Investigator(Kenkyū-buntansha): 篠原 歩 東北大学, 情報科学研究科, 教授 (00226151) ; 正代 隆義 九州国際大学, 国際関係学部, 教授 (50226304) ; 畑埜 晃平 九州大学, 附属図書館, 准教授 (60404026) ; 吉仲 亮 京都大学, 情報学研究科, 助教 (80466424) ; 内沢 啓 山形大学, 理工学研究科, 准教授 (90510248)
• Research Collaborator: TSUDA Koji ; CUTURI Marco
• Project Period (FY): 2012-06-28 – 2017-03-31
• Keywords: 計算学習理論 / オンライン予測 / 計算理論 / 分布学習

Outline of Final Research Achievements

We made considerable achievements in bringing new insights into "computation", through the analysis of the complexity of learning problems. In the analysis of online prediction, we partially solved an important open problem, stating the equivalence between the complexity of online prediction and that of optimization. In the analysis of the complexity of hypothesis representation, we proposed a few decision problems defined on the discrete dynamical system, which is a generalization of recurrent neural networks. Our analysis suggests that the decision problems belong to an intermediate layer of the class of NP. In the analysis of learning grammars, we employs a breakthrough technique called the distributional learning and obtained many positive results in learning formal (graph) grammars. The distributional learning is a resume of designing learning algorithms by focusing on the relation of substrings/subgraphs and the structures surrounding them.

Free Research Field

計算学習理論，計算理論，アルゴリズム