Neural network engine
A neural network engine uses a trained neural network to evaluate positions, and NNUE, the efficiently updatable neural network added to Stockfish, combines such an evaluation with fast traditional search.
Also called: NNUE
A neural network engine judges positions using a network that has been trained on very large numbers of positions and results, rather than relying only on hand-written rules about material and structure. The network learns patterns that are hard to describe in words, such as how a certain pawn structure and piece placement tend to favor one side.
NNUE stands for efficiently updatable neural network. It is a compact network designed so that when a move changes only a few squares, most of the calculation can be reused instead of repeated. This makes it fast enough to run on an ordinary processor inside a traditional alpha-beta search. Stockfish adopted NNUE and became noticeably stronger as a result, and other engines have used similar ideas. Some other engines are built around larger networks and a different style of search.
For a student, the lesson is that modern engine evaluations are a blend of learned judgment and deep calculation. Their verdicts about long-term factors such as space, activity and king safety are now often more accurate than before. Still, an engine does not explain itself, so use its answers as a prompt for your own understanding.
An engine can tell you what "Neural network engine" means for a position, but not which mistakes you keep making; a first lesson with George Li reviews your games and finds those.
Related terms
Stockfish is a free, open-source chess engine developed by a worldwide community, widely regarded as one of the strongest engines available and used by many analysis websites and apps.
A chess engine is a computer program that searches positions and evaluates moves, usually playing and analyzing far stronger than any human, and it is used for analysis, preparation and play against a machine.
An evaluation is a numerical judgment of who stands better in a position, usually given in pawns from White’s point of view, where positive favors White and negative favors Black.
Alpha-beta pruning is a search technique that skips branches of the move tree that cannot change the minimax result, letting engines look much deeper using the same amount of computing time.
Search depth is how many moves ahead an engine has looked in a position, counted in half-moves called plies, and deeper searches generally give more reliable evaluations and moves.
Engine analysis is the use of a chess program to search a position and report which moves are strongest, along with a numeric evaluation and the lines it expects both sides to follow.
Guides that go further
What to decide first, the questions to ask, what a first lesson should show and the red flags to avoid.
How much does chess coaching cost?
Published rates for private, group and subscription chess coaching, with the source and the date of each.
A game review, a read on your level and a written study plan from George Li.
Want to spot this in your own games?
A first lesson starts with your games: George reviews them and builds a study plan around what he finds.