I am a computer scientist based in Tokyo, Japan. Currently, I am managing a research group on natural language processing π, data engineering ποΈ, and acoustics π£οΈ research field at NEC Corporation as the π§βπΌ Head of Group and a π§βπ¬ Research Fellow . I received my Ph.D. at University of Tsukuba in March 2018 under supervision of Prof. Hiroyuki Kitagawa.
My research interests are on how people can unleash their full potential using the power of machines π€ and knowledges π. Specifically, my current research interests are on understanding the behavior of large language models (LLMs), semi-parametric natural language methods based on external corpora retrievers (such as augmented language models), autonomous knowledge management using LLMs, and language model cascades (agent LLMs).
Journal of Information Processing Outstanding Paper Award, 2021
IPSJ
IPSJ Yamashita SIG Research Award, 2019
IPSJ
Best Paper Runner-Up, 2018
WebDB Forum
Annual Conference Award, 2016
JSAI
Best Honorable Poster, 2015
DEIM
Best Paper Runner-Up, 2015
APWeb
Honorable Poster, 2014
DEIM
Dept. Chair Award, 2013
University of Tsukuba
Honorable Student Talk, 2013
DEIM
PhD in Computer Science, 2018
University of Tsukuba
MEng in Computer Science, 2013
University of Tsukuba
BSE in Computer Science, 2011
University of Tsukuba
Our research team (knowledge-based learning) at NEC Corporation is seeking for motivated full-time researchers / internship students who are passionate in working on interdisciplinary research issues that arise from real-world enterprise business. We aim to contribute to both industries and academics. Our research results are commercialized and used in various enterprise companies such as retailers and consumer products companies. We publish our results in top-level venues of computer science (e.g., AAAI, ICDE, ICDM, BigData).
Research topics include but not limited to:
Please drop me an E-mail with your CV if you are interested in working with us.
See my GitHub for list of software I have crafted. Here are some selected stuffs.
*Provides fully-configurable hardware keyboard functionalities for web browsing on iOS (iPadOS)
Yet another keyboard remapping tool for X environment
An interactive grep tool in your terminal
Allows you to bind commands to key sequences in Mozilla Firefox
A Lisp implementation and REPL written in JavaScript, which supports static-scoping, lexical-closure, macro, and basic special forms
A parser for org-mode notation written in JavaScript
Provides zsh like completion for minibuffer in Emacs
A chaos fractal generator written in JavaScript
Add jsdoc-related functionalities for Emacs
A major-mode for LEMON Parser Generator.
Complete list of publications is also available.
Many annotation systems provide to add an unsure option in the labels, because the annotators have different expertise, and they may not have enough confidence to choose a label for some assigned instances. However, all the existing approaches only learn the labels with a clear class name and ignore the unsure responses. Due to the unsure response also account for a proportion of the dataset (e.g., about 10-30% in real datasets), existing approaches lead to high costs such as paying more money or taking more time to collect enough size of labeled data.
Designing good features for machine learning models, which is called feature-engineering, is one of the most important tasks in data analysis. Well-designed features, which capture the characteristics of data, improve the predictive performance and explainability of the model. Since good features generally reflect the deep knowledge on business domains of the data and the analysis task, feature engineering is considered as one of the most difficult phases in data analysis.
Given a large amount of table data, how can we find the tables that contain the contents we want? A naive search fails when the column names are ambiguous, such as if columns containing stock price information are named βCloseβ in one table and named βPβ in another table. One way of dealing with this problem that has been gaining attention is the semantic annotation of table data columns by using canonical knowledge.
Range aggregation query is a fundamental operation in the feature engineering phase of the machine learning tasks, which computes statistics, such as the maximum and the standard deviation of a subset of records. Since the feature-engineering process is a trial-and-error process, data analysts repeatedly conduct tons of the range aggregation queries by changing the range conditions, which results in a heavy workload. To accelerate such repetitive range aggregation queries, we propose Adaptive Partial Aggregation Tree (APA-tree), which drastically reduces the amount of I/Os that happen in executing the range aggregation queries.
In this paper, we present CVS (Compressed Vector Set), a fast and space-efficient data mining framework that efficiently handles both sparse and dense datasets. CVS holds a set of vectors in a compressed format and conducts primitive vector operations, such as lp-norm and dot product, without decompression. By combining these primitive operations, CVS accelerates prominent data mining or machine learning algorithms including k-nearest neighbor algorithm, stochastic gradient descent algorithm on logistic regression, and kernel methods.
Given a collection of basic customer demographics (e.g., age and gender) and their behavioral data (e.g., item purchase histories), how can we predict sensitive demographics (e.g., income and occupation) that not every customer makes available? This demographics prediction problem is modeled as a classification task in which a customer’s sensitive demographic y is predicted from his feature vector x. So far, two lines of work have tried to produce a “good” feature vector x from the customer’s behavioral data: (1) application-specific feature engineering using behavioral data and (2) representation learning (such as singular value decomposition or neuralembedding) on behavioral data.
This paper presents a new probabilistic generative model (PGM) that predicts links for isolated nodes in a heterogeneous network using textual data. In conventional PGMs, a link between two nodes is predicted on the basis of the nodesβ other existing links. This method makes it difficult to predict links for isolated nodes, which happens when new items are recommended. In this study, we first naturally expand the relational topic model (RTM) to a heterogeneous network (Hetero-RTM).
A recent trend in data stream processing shows the use of advanced continuous queries (CQs) that reference non-streaming resources such as relational data in databases and machine learning models. Since non-streaming resources could be shared among multiple systems, resources may be updated by the systems during the CQ execution. As a consequence, CQs may reference resources inconsistently, and lead to a wide range of problems from inappropriate results to fatal system failures.
Principal investigator of a research team (knowledge-based learning). Research topics include
Research on customer behavior data analytics. Research topics include