Shuhei M. YOSHIDA, PhD

Researcher in machine learning and computer vision

Visual Intelligence Research Laboratories, NEC Corporation

I am a researcher in machine learning and computer vision at Visual Intelligence Research Laboratories, NEC Corporation. My main research interest is in the field of machine learning algorithms that work with limited supervision, such as weakly supervised learning and transfer learning.

I got my Ph. D. in physics at The University of Tokyo. The theme of my Ph. D. study was universality in strongly correlated quantum gases and few-body systems.

“Shuhei Yoshida” is such a common name that I once met a guy with the name in an physics conference! Please don’t confuse me with other “Shuhei Yoshida”. Most of my publications in English have the middle initial “M” in my name.

Interests

Machine learning with limited supervision (weakly supervised learning, transfer learning)
Deep learning algorithms, theories, and their application to image/video recognition

Education

PhD in Science, 2018
Graduate School of Science, The University of Tokyo
MSc, 2015
Graduate School of Science, The University of Tokyo
BSc, 2013
School of Science, The University of Tokyo

Featured Publications

Action-Agnostic Point-Level Supervision for Temporal Action Detection

We propose a novel form of supervision called action-agnostic point-level (AAPL) supervision for temporal action detection. Extensive experiments demonstrate that the proposed approach is competitive with or outperforms prior methods in terms of the trade-off between the annotation cost and detection performance.

Shuhei M. Yoshida, Takashi Shibata, Makoto Terao, Takayuki Okatani, Masashi Sugiyama

DOI arXiv

Lower-Bounded Proper Losses for Weakly Supervised Classification

We theoretically derive conditions under which loss functions for weak-label learning are proper and lower-bounded – two essential requirements for the class-probability estimation. This theorem leads us to a novel regularization called generalized logit squeezing, which empirically outperforms improper or unbounded losses.

Shuhei M. Yoshida, Takashi Takenouchi, Masashi Sugiyama

ICML virtual site (Registered user only) Proceedings arXiv

See all publications