# TAKENOUCHI, Takashi

Department | Department of Complex and Intelligent Systems |
---|---|

Specialized Fields | Machine learning, Pattern recognition. |

Subjects in Charge | Probability and Information Theory, Complex Computation Theory, Mathematics Practice 1,2 |

Academic Background | The Graduate University for Advanced Studies |

Degree | Doctor of Philosophy |

Personal History | 1999.3 University of Tokyo(B.Eng.) 2001.3 University of Tokyo(M.Eng.) 2004.3 the Graduate University for Advanced Studies(Ph.D.) 2004.4 Institute of Statistical Mathematics,Project Researcher 2005.4 Nara Institute of Science and Technology, Bioinformatics Unit, Postdoctoral fellow 2006.4 Nara Institute of Science and Technology, Laboratory for Theoretical Life-Science, Postdoctoral fellow 2008.4 Nara Institute of Science and Technology, Laboratory for Theoretical Life-Science, Assistant professor 2011.4 Nara Institute of Science and Technology, Mathematical Informatics Laboratory, Assistant professor |

Starting Time of Employment | 2012.4 |

## Research Contents

I am interested in Machine learning and Patter recognition.

・ Machine learning: Human can learn by experience and modify own action based on the knowledge. The Machine learning constructs systems to achieve the learning process and to extract useful information from a dataset, using mathematical models and various kinds of statistical techniques.

・ Pattern recognition: Pattern recognition is a problem which assigns a label to a given input value. To achieve higher prediction accuracy, a more complex learning machine requiring heavy computational cost is necessary. An ensemble learning method is a promising approach to avoid the problem of heavy computational cost without loss of generalization ability. Like the proverb ``Two heads are better than one”, the method combines numbers of simple learning machines which can be learned with low computational cost, to improve its performance.

## Attractive Factors of My Research

An essence of the machine learning and pattern recognition is to extract information such as relation between the input and the output, from the dataset. While human can easily do it for a small scale dataset, it is very difficult to extract information when the dataset becomes large-scale and complex. We construct mathematical models to tackle the large-scale and complex dataset and theoretically investigate performance of models. The most fascinating point of the machine learning is that the mathematical model enables us to conduct advanced information processing beyond human ability.

## Achievements

I have proposed various kinds of methods for pattern recognition and theoretically investigated performance of methods.

Robustification: Pattern recognition method is basically learned with the label information of input value. The label is basically assigned by human decision and then can be contaminated by noise of mislabeling. To tackle with the problem, I proposed a probabilistic model of mislabeling which leads to a robust pattern recognition method.

Maximization of AUC criteria: In the area of medical diagnosis, maximization of AUC criteria rather than the classical misclassification error is important. I proposed a method which can directly maximize AUC criteria, and provided theoretical analyses which ensure virtues of the method such as consistency and robustness.

## Major Books and Papers

A novel boosting algorithm for multi-task learning based on the Itakura-Saito divergence.

In Proceedings of Bayesian Inference And Maximum Entropy Methods In Science And Engineering (MAXENT 2014), pp. 230—237, 2014.

An extension of the Receiver Operating Characteristic curve and AUC-optimal classification.

Neural computation, 24(10), pp. 2789-2824, 2012.

Ternary Bradley-Terry model-based decoding for multi-class classification and its extensions.

Machine Learning, 85(3), pp.249-272, 2011.

A multi-class classification method based on decoding of binary classifiers.

Neural Computation, 21(7), pp.2049-2081, 2009.

## Message to Students

Machine learning is an essential framework in the area requiring prediction techniques. Please join my lab and enjoy research with mathematical modeling.