Computer vision and multimedia :Toshihiko Yamasaki

Attractiveness Computing

We are interested in analyzing why and how we are attracted to specific persons, content, and services. We have been trying to analyze, tell reasons, and even enhance such “attractiveness” in multimedia big data. We are not doing research on application oriented topics, but trying to solve fundamental research problems behind them.

Exploring new areas in machine learning and pattern recognition

We have been working on novel machine learning algorithms, not simple extensions of existing algorithms.

・Learning with few/imperfect training data
We are working on hierarchical transfer learning, unpaired learning, weakly-supervised learning, contrastive learning, and so on for robust and practical applications.

・Reliable learning
We have been working on understanding mechanisms and its defense of adversarial examples and fooling images. Besides, we are working on fake image/video detection.

・Reinforcement learning and meta learning
We have been working on deep learning based photo/video procesing. In particular, we are interested in reinforcement learning for image editing, filtering, video summarization and so on.

Real-world applications and others

We are also challenging new research topics aiming at widening our research activities.

・Medical image analysis

・Tools for illustration drawing and CG generation

・Nursery school and elderly care house sensing

・Environment sensing using our own IoT devices

・Deep learning in severe env. such as space

・Action recognition and retrieval

・Fundamental CV problems such as feature point matching, inpainting, super-resolution, etc.