27 July 2020

A Chess Monsterpiece?

In last week's post, Stockfish Wins TCEC S18; Leela Wins CCC14, I finished saying,

The CCC then conducted a series of 'Post-CCC14' matches, most of them featuring a new 'Stockfish NNUE' engine. Since I didn't have the time to look at these matches in any detail, I'll save that for the next, off-week post in the TCEC/CCC series.

A mid-July announcement from the CCC, Stockfish NNUE, Strongest Chess Engine Ever, To Compete In CCCC (chess.com), said,

Stockfish NNUE has broken new ground in computer chess by incorporating a neural network into the already incredibly powerful Stockfish chess engine. This hybrid chess monsterpiece is estimated to perform better than the current Stockfish 11 who recently defeated Leela Chess Zero in the TCEC, giving it a solid claim to being the strongest chess engine in the world.

Monsterpiece? A CCC command informs,

!nnue • NNUE (Efficiently Updatable Neural Network) is a new class of neural network based nonlinear evaluation functions originally designed for computer shogi. NNUE evaluation functions are designed to run efficiently on CPU using various acceleration techniques, including incremental computation.

Further information, including a long discussion of the new engine, is at Stockfish NN release - NNUE (talkchess.com; May 2020). It starts,

A year ago somebody ported Shogi NN called NNUE (Efficiently Updateable Neural Network backwards) to SF10 as a proof of concept.

A first CCC trial didn't go well for the newcomer. Lc0 beat Stockfish NNUE in a 200 game match by a score of +32-11=157 (110.5-89.5) at a time control of 3'+2". A second trial using the same time control added plain vanilla Stockfish to the two other engines. That trial is still underway.

Getting back to the name, why not just call the engine Neural Network Updateable Efficiently? The phrase 'backwards' often means 'bass ackwards'.

No comments: