Ashish Behal
AI / ML Engineer
← All Projects

Chess
Diffusion

Trained a Discrete Denoising Diffusion Probabilistic Model (D3PM) on chess puzzle positions encoded as 72-token FEN sequences — with no explicit knowledge of the rules. The model achieves 69.5% valid position generation versus a 16.3% random baseline, and closely matches the training distribution on pawn structure metrics.

69.5%
Valid positions generated
4.3×
The random baseline rate

Features

Results

The model excels at learning pawn structure — closely matching the training distribution on white pawn count and passed pawn metrics. Material balance and total material are learned less precisely, with the model showing a systematic bias toward piece-rich opening positions.

White Pawn Count Distribution

White Pawn Count Distribution

Passed Pawn Count Distribution

Passed Pawn Count Distribution

Material Balance Distribution

Material Balance Distribution

Total Material Distribution

Total Material Distribution