Haruo HOSOYA

Senior Researcher Department of Dynamic Brain Imaging Brain Information Communication Research Laboratory Group ATR Institute International Email:

Current Research: Computational neuroscience and Brain-inspired AI

I develop brain-inspired AI models that leverage computational principles uncovered through neuroscience, with a particular focus on the brain’s powerful ability to generalize across tasks and contexts. My research aims to identify the mechanisms that support such flexibility—especially in the visual cortex and hippocampal formation—and translate them into AI architectures that go beyond narrow, task-specific performance. By integrating insights from perception and memory, I seek to build models that are not only efficient and scalable but also capable of robust, human-like generalization. This cross-disciplinary approach advances our understanding of brain function while paving new paths toward more adaptive and intelligent AI systems.

Keywords

On early/intermediate vision:

• AI/ML: hierarchical sparse coding, natural image statistics, drastic dimension reduction, Bayesian inference
• Neuroscience: shape coding

On higher vision:

• AI/ML: mixture of sparse coding, group-based weakly supervised learning (VAE-based), categorical invariant representation, disentanglement, one-shot classification, invariant clustering
• Neuroscience: category-specific object coding, face-processing patches

On memory system:

• AI/ML: memory-augmented model, abstraction, relational structure, very-few-sample learning, domain transfer, analogy
• Neuroscience: conceptual grid cells

Publication list

Research Outline

Biography

• 2001 PhD, Dept. of Computer Science, University of Tokyo.
• 2001-2003 Assistant Professor, RIMS, Kyoto University.
• 2003-2011 Lecturer, Dept. of Computer Science, University of Tokyo.
• 2011-2012 JST Presto Researcher, BSI, Riken.
• 2012-2016 JST Presto Researcher, ATR.
• 2016-present Senior Researcher, ATR.
• 2023-present Specially appointed researcher (adjunct), University of Tokyo.

Previous Research

Many people are astonished, but I'm originally from "hard-core" computer science. I used to work on applying discrete mathematics such as automata, logic, and type theory to the design and implementation of novel programming languages in the direction of helping programmers in safety, productivity, and efficiency. My main target in this research direction was to use theory of tree automata to the design of type-safe programming languages specialized to processing XML data. The project ended in 2010.

Book

Foundations of XML Processing: The Tree-automata approach (Cambridge University Press)

Personal Activities

I have been enjoying since my childhoold playing European classical music with the piano. My recent favorites are French modern pieces, especially, Debussy and Ravel.

Here are some of my performances on YouTube.

I won a few prizes at competitions:

• the third place at Ikaruga music competition in 2016 (general superior category).
• the fourth place at Nakatsu An die Musick competition in 2019 (general category).
• finalist at Osaka International Music Competition in 2020 (section I-G).
• the forth place at France Piano Competition in 2022 (general category).
• the first place, Debussy award at Debussy International Competition in 2022 (amateur category).
• finalist at Japan Classical Music Competition in 2022 (general category).
• the third place at Japan Chamber Music Piano Competition in 2024 (amateur category).

My favorite quotes

"... the process of acceptance will pass through the usual four stages:

1. This is worthless nonsense.
2. This is an interesting, but perverse, point of view.
3. This is true, but quite unimportant.
4. I always said so."

--- J.B.S. Haldane (1963)