Draft:Yu Nasu
Japanese computer scientist
From Wikipedia, the free encyclopedia
Yu Nasu is a Japanese computer scientist and shogi programmer, widely recognized for developing the NNUE (Efficiently Updatable Neural Network or ƎUИИ) evaluation function.
Comment: There appear to still be some issues with the references not verifying the statements, which is common with LLM. In addition notability is not dhown either as a scientist or general. Finally the format is very wrong in nany places, please read WP:MOS. Ldm1954 (talk) 01:49, 1 March 2026 (UTC)
 | This is a draft article. It is a work in progress open to editing by anyone. Please ensure core content policies are met before publishing it as a live Wikipedia article. Find sources: Google (books · news · scholar · free images · WP refs) · FENS · JSTOR · TWL Last edited by Ldm1954 (talk | contribs) 15 days ago. (Update)
Finished drafting? |
Yu Nasu | |
|---|---|
那須 悠 | |
| Alma mater | Tokyo Institute of Technology |
| Occupations | Computer scientist, Shogi programmer |
| Employer | Toshiba (Knowledge Media Laboratory) |
| Known for | NNUE (Efficiently Updatable Neural Network) |
| Notable work | YaneuraOu (NNUE implementation), Tanuki |
Early Life and Education
Yu Nasu (Japanese: 那須 悠) pursued his higher education at the Tokyo Institute of Technology (TITech), a leading institution for engineering and computer science in Japan. During his time at the university, he was affiliated with the Shinoda-Furui Laboratory, where he worked under the supervision of prominent speech recognition experts Koichi Shinoda and Sadaoki Furui.[1]
His academic training focused heavily on Hidden Markov model (HMM) and statistical signal processing. In 2011, as part of his research at TITech, Nasu co-authored work on cross-channel spectral subtraction, specifically targeting the improvement of speech recognition accuracy in complex environments such as meetings. His studies at the institute provided the foundational expertise in neural networks and incremental computation that would later inform his breakthroughs in computer shogi. Following his graduate studies, he transitioned into the private sector, joining Toshiba Corporation, where he continued his research at the company's Knowledge Media Laboratory.[2]
Career and Research
After graduating from the Tokyo Institute of Technology (TITech), Nasu joined Toshiba Corporation as a researcher at its Knowledge Media Laboratory. His research during this period focused on speech technology and human–computer interaction, with an emphasis on statistical speech synthesis.[3]
Nasu's work involved developing techniques to increase the naturalness and expressiveness of machine-generated voices. He notably co-authored research on "emotional transplant" techniques, which utilized an emotion additive model (EAM) to allow speech synthesis systems to adopt specific emotional tones from one speaker to another.[4]
Development of NNUE
In 2018, Nasu introduced the Efficiently Updatable Neural Network (NNUE),[5] a shallow neural network architecture specifically designed to replace traditional handcrafted evaluation functions in computer shogi. The name is a Japanese wordplay on Nue, a mythical chimera, and is often stylized in technical documents as ƎUИИ. Unlike the deep convolutional neural networks used by systems such as AlphaZero, which require significant TPU resources, Nasu's architecture was optimized for high-speed execution on standard CPUs.[6]
Technical architecture and efficiency
The core innovation of NNUE lies in its incremental update mechanism. Most of the network's knowledge is stored in the first layer, which uses a large but sparse input vector, typically over 40,000 boolean inputs in shogi, representing piece-square relationships relative to the king. Nasu's design exploits the fact that a single move only changes a few board pieces. Instead of recomputing the entire network, the system maintains an accumulator that updates only the affected neurons. This allows the engine to evaluate millions of positions per second, maintaining the high search speeds required for alpha-beta pruning.[7][2]
Implementation and global impact
Nasu first implemented NNUE in a modified version of the YaneuraOu shogi engine, which immediately demonstrated superhuman playing strength on par with deep-learning-based systems. In 2020, Japanese programmer Hisayori "Nodchip" Noda successfully ported Nasu's architecture to the international chess engine Stockfish. This adaptation, released as Stockfish 12, resulted in an unprecedented increase of approximately 80-100 Elo points, marking the most significant jump in the engine's history. Since then, NNUE has became the industry standard for nearly all top-tier chess and shogi software.[8]
See also
- Stockfish (chess): The world-leading chess engine that utilizes NNUE as its core evaluation function.
- Computer shogi: The field of computer science where Yu Nasu's innovations originated.
- Neural network (machine learning): The broader machine learning technology upon which NNUE is built.
- Alpha–beta pruning: The search algorithm that NNUE was specifically designed to complement.