Tu banner alternativo

Efficiently updatable neural network

In this article, the topic of Efficiently updatable neural network will be addressed, which has generated great interest and controversy in recent times. Efficiently updatable neural network is a topic that has captured the attention of people of all ages and backgrounds, since its relevance transcends borders and contexts. Since its emergence, Efficiently updatable neural network has sparked much debate and has been the subject of study and analysis by experts and hobbyists alike. In this article, different aspects related to Efficiently updatable neural network will be explored, including its origin, evolution, impact and possible future implications. Likewise, different perspectives and opinions on Efficiently updatable neural network will be examined, with the aim of offering a complete and balanced view on this topic.

Tu banner alternativo
This article is part of the series on
Chess programming
Board representations
Evaluation functions
Tuning and training algorithms
Graph and tree search algorithms
Chess computers
Chess engines

In computer strategy games like shogi and chess, an efficiently updatable neural network (NNUE, a Japanese wordplay on Nue, sometimes stylised as ƎUИИ) is a neural network-based evaluation function whose inputs are piece-square tables, or variants thereof like the king-piece-square table.[1] NNUE relies on the tendency in these games for the game state to change only slightly between moves.

NNUE was invented by Yu Nasu and introduced to computer shogi in 2018.[2][3] On 6 August 2020, NNUE was for the first time ported to a chess engine, Stockfish 12, resulting in a major increase in playing strength for that engine.[4][5]

NNUE are designed to run efficiently on central processing units (CPU).[6][7] They use incremental computation and single instruction multiple data (SIMD) techniques along with appropriate intrinsic instructions.[2] In contrast, deep neural network-based chess engines such as Leela Chess Zero require GPU-based inference.[8][9]

The neural network used for the original 2018 computer shogi implementation consists of four weight layers: W1 (16-bit integers) and W2, W3 and W4 (8-bit). It has 4 fully-connected layers, ReLU activation functions, and outputs a single number, being the score of the board. As of 2025, Stockfish has introduced several optimizations to the NNUE architecture, but the overall architecture remains similar.[10]

See also

References

  1. ^ Gary Linscott (April 30, 2021). "NNUE". GitHub. Retrieved December 12, 2020.
  2. ^ a b Yu Nasu (April 28, 2018). "Efficiently Updatable Neural-Network-based Evaluation Function for computer Shogi" (PDF) (in Japanese).
  3. ^ Yu Nasu (April 28, 2018). "Efficiently Updatable Neural-Network-based Evaluation Function for computer Shogi (Unofficial English Translation)" (PDF). GitHub.
  4. ^ "Introducing NNUE Evaluation". 6 August 2020.
  5. ^ Joost VandeVondele (July 25, 2020). "official-stockfish / Stockfish, NNUE merge". GitHub.
  6. ^ "Stockfish FAQ: Can Stockfish use my GPU?". Stockfish. Retrieved 19 January 2025.
  7. ^ "nnue-pytorch/docs/nnue.md". GitHub.
  8. ^ Dominik Klein (2022), Neural Networks for Chess, p. 49, arXiv:2209.01506
  9. ^ Monroe, Daniel; Chalmers, Philip A. (2024). "Mastering Chess with a Transformer Model". arXiv:2409.12272 .
  10. ^ "NNUE | Stockfish Docs". Stockfish documentation. 2025-06-12. Retrieved 2025-06-23.
Wikious
Boobota
Sagapedia