Research and Education Fields

We have been working on improvement of very accurate classifiers. Especially, we are considering a system consisting of obtaining samples, extraction of features, slection of features, and construction of classifiers. (Handwritten character recognition, biometrics human identification, recognition of human taste, medical diagnosis, and so on)[Details]

Design of classifiers

Volume prototypes

Fast probably correct (PC) k-nearest neighbor classifiers

Feature Selection

Multi-label Classification

Keywords: Classifiers, k-Nearest Neighbor, Feature Selection, Volume Prototypes, Multi-Label Classification

Imbalance problems

Unlike biometrics identification (“hard authentication”), we are consdiering “soft authetication” using sensors whose existence is unawared to users. Soft authentication is different from traditional hard authentication even in the goal. The former is mainly for keeping security but the latter is for giving personalized service in home environment or a small office environment.(Person tracking by ultra-red ceiling sensors: Human identification by pressure sensors attached on a chair, and so on)[Details]

Elves IT

User Tracking

User Authentication without Cooperative Action

Health monitoring of a sitting person

Fall Detection and Prevention

Measurement group sleepiness

Keywaords: Fairy IT, Soft Authentication, Fall Detection, Sleepiness Ditection

There are various relations between persons, between things and between a person and a thing, and those can be represented as a form of a network. Social networks and communication networks are also considered as representations of certain relations between persons and between things. Through network analysis, some useful knowledge might be obtained from the relations between them. Our study mainly focuses on structures and time-series changes of the networks that exist over the internet, such as social networks.[Details]

Community Extraction

Anomaly Detection

Trend Tracking

Influence Maximization

Keywaords: relational data analysis, anomaly detection, time-series of graphs, tensor factorization, influence maximization

The wide spread of internet has brought easiness of interaction between a user and a system. Through such interaction, the system learns user’s preference and provides better services. We have been studying ways of online selection of information to deriver, which are effective in ad delivery and recommendation. [Details]

Bandit Problem

Recommendation

Keywords: online learning, machine learning, learning theory

With the progress of genome projects, huge nucleotide sequences of chromosomes of various animals and plants have been made known. With the development of computers and communication technology, various time-series data have been automatically recorded and stored. We are developing string algorithms to analyze such huge sequences. We mainly focus on similar structures that iteratively appear in a sequence. [Details]

Analysis of Repetition Structure

Keywords: tandem repeat, interspersed repeat