In iTHEMS Biology Seminar on November 20th, Dr. Hidenori Tanaka (Physics & Informatics Laboratories, NTT Research) gave us an exciting talk about physics principles in neural networks. He first reviewed the basic scheme of deep learning using neural networks. Then, he presented three questions regarding both neural science and machine learning and explained his recent works which address these questions. He stressed how physics principles like symmetry and conservation laws are useful in extracting minimal features of biological circuit models, improving algorithms to simplify neural networks, and predicting learning dynamics of neural networks. As his talk was clear and kind to both specialists and non-specialists, there were various questions from the audience. Hidenori is a very active researcher, and I was happy to invite him as a guest speaker.

Kyosuke Adachi (BDR/iTHEMS)