Rikiya Takahashi (高橋 力矢)

Contact Information IBM Research, Tokyo Research Laboratory 1623-14 Shimotsuruma, Yamato-shi, Kanagawa-ken 242-8502 Japan e-mail:  I am a researcher of data analytics and statistics in Tokyo Research Laboratory (TRL) since 2004. Currently I belong to Data Analytics Team in Business Services Research group. My interests are i) to model many kinds of complex data caused by complicated human behaviors, and ii) to derive practical business insights such as marketing expertise, using statistics and optimization techniques. In research perspective, I am interested in time-series modeling, prediction, classification or clustering. Especially, my challenge is to robustly and efficiently deal with the sequences in which long-term dependencies and bustiness exist. These long-term dependencies are observed in many time-series data including natural language, customer purchase history, and network packets, but standard stochastic models such as Markov models are insufficient to model them. Because of previous study in automatic speech recognition and natural language processing, I am already familiar with many probabilistic models such as high-order visible or hidden Markov models. My current work is to develop more advanced stochastic models to efficiently model long-term dependent sequences, such as Dirichlet-compound-multinomial models. In business perspective, I work in marketing research and risk management. In the marketing research area, recently I am fascinated by psychological approaches, which are sometimes quite different from normal statistical approaches. Several psychological experiments based on questionnaires can clarify interesting criteria of humans decisions and their irrationality. These results often imply inefficiency of adopting standard predictive models. Dealing with this irrationality and deriving practical insights from these experiements will be academically interesting and be an essential area of Service Science. Education M.S. in Information Science and Technology, 2004, University of Tokyo, Tokyo, Japan B.Eng. in Information and Communication Engineering, 2002, University of Tokyo, Tokyo, Japan Conference Papers, Referred R. Takahashi, “Separating Precision and Mean in Dirichlet-Enhanced High-Order Markov Models”, In proceedings of The 18th European Conference on Machine Learning (ECML 2007), pp.382-393, 2007 Other Papers R. Takahashi, T. Osogami, “Action-Sensitive Hidden Markov Model of Customer Behavior”, In proceedings of New Directions of Research in Marketing (University of Tsukuba and Tohoku University), pp.119-134, 2006 R. Takahashi, N. Minematsu, K. Hirose, “Enhancing accuracy of dynamically interpolating language model using multiple backing-off N-grams”, In proceedings of Special Interest Group of Spoken Language Processing (Information Processing Society of Japan), 49(11), pp.61-66, 2003 R. Takahashi, N. Minematsu, K. Hirose, “A dynamically context-adapted language model using multiple N-grams interpolation”, In proceedings of Special Interest Group of Spoken Language Processing (Information Processing Society of Japan), 46(7)，pp.37-42, 2003 Others Privately I am an amateur composer of classical musics, using strings and a piano. I seem to prefer the scores of late romantic age, and I admire Arnold Schoenberg, Gustav Mahler and Jean Sibelius. I also adore Ludwig van Beethoven as a great composer. I also like film musics and my favorite composer in film music area is definitely Jerry K. Goldsmith ( Especially, "Papillon", "QBVII", and "Firstblood" are incredible!! ). Following score is an excerpt from a variation I composed. This score is written for my practice. In vacations, I prefer to visit counties that have clystalline environments, such as North Europe. My teacher of music composition says that my preference in travels seems to strongly affect the music I compose, though I am not so aware. Hallstatt in Austria Gullfoss in Iceland Jökulsárlón in Iceland