Satoshi Imamura

Master's student in Electrical Engineering and Computer Science at Tokyo University of Agriculture and Technology. I work on representation learning for EEG, and build small open-source tools on the side.

hold still Hold still for a moment and the print develops. Move, and it fades back to paper. Dithered with ditherpress.

Research

Toshihisa Tanaka Lab, Tokyo University of Agriculture and Technology

Representation learning for EEG

Master's research · Apr 2026 – present

Learning general-purpose representations of brain signals that transfer across tasks and datasets.

An EEG foundation model with vector quantization

Bachelor's thesis · Mar 2026

Built and evaluated an EEG foundation model that turns continuous signals into discrete tokens through a learned VQ codebook.

Automatic sleep-stage classification

Research internship · 2024

Compared sleep-stage classifiers on EEG, EOG and EMG with MNE-Python and YASA. Notebooks

Projects

sasara

TypeScript · npm

A Japanese bunsetsu-level dependency parser. A ~550 KB linear model, no neural network, runs in Node.js and the browser, and reports a calibrated confidence for each attachment.

ditherpress

Python · single file

Dithers photos and short video loops down to a few flat colours, keeping the dots still between frames so the animation doesn't shimmer. The print at the top of this page is its output.

id-photo-maker

React · TypeScript · MediaPipe

Lays out ID photos on an L-size sheet you can print at a Japanese convenience store for about ¥30. Everything runs in the browser.

monokai-pro.wezterm

Lua

Monokai Pro for WezTerm, with all six filters.

About

I grew up between Japan and China and work in Japanese, Chinese and English. Outside the lab I take photos, sing a cappella and beatbox. Day to day I live in Neovim and WezTerm.

日本語: native
中文: native · HSK 6
English: professional · TOEIC 970