Doomformer Turns DOOM Maps Into Neural Networks That Recognize Handwritten Digits
Developer mottopanikeiku released Doomformer, a project that converts an integer neural network into a DOOM map and runs the arithmetic inside the game.
1 source 1m ago
Reporting from 1 source: GIGAZINE.
Why it matters
The project treats DOOM's map logic as a general computing substrate rather than a game, which puts a working Transformer-style attention step inside a 1993 shooter's engine.
Key Facts
- Developer mottopanikeiku released Doomformer, a project that compiles an integer neural network into a DOOM map.
- Doomformer uses voodoo dolls, player clones created by multiple player starts, to perform the arithmetic.
- Calculation results are stored as light levels and as floor and ceiling heights.
- The handwritten digit demo uses 1,355 clones and recognized the correct digit in all 30 developer-prepared trials.
- Doomformer recorded 97.16% accuracy on 10,000 MNIST test images.
Voodoo dolls, the player clones created by multiple player starts, move along conveyor-like floors; light levels and floor and ceiling heights store the values. In a demo it recognizes handwritten digits, scoring 97.16% on 10,000 MNIST test images.
Doomformer targets ZDoom-family source ports, and the developer says every value computed inside the map was checked against NumPy with a tolerance of 0. The demo's digit recognition uses 1,355 clones and reports the correct digit in all 30 trials where the developer wrote each digit 0 through 9 in three different ways. Attention, the mechanism used in Transformers, is reproduced by controlling clone passage with ceiling heights.
Written with AI by Yomimono from the 1 cited source below, including Japanese-language reporting where cited. See our editorial policy.
Add Yomimono as a preferred source on Google