
MNIST64
Interactive 6502-based implementation of a 99.5% accuracy binarized neural
1 ratings4.19
Saves2
CategoryAI
PlatformWeb
PricingFree
MNIST64 is an extremely optimized, interactive binary neural network running on a stock Commodore 64 computer. It achieves near-state-of-the-art handwritten digit recognition accuracy of 99.5% on MNIST+EMNIST datasets using custom 6502 assembly code without multiplication instructions. The project includes a fully working software product validated on original hardware, featuring early-exit cascades and test-time augmentation for adaptive speed and accuracy.
Ratings & Reviews
Write a reviewBe the first to share a thoughtful review.
Information
CategoryAI
PlatformsWeb
PricingFree
AccountNot confirmed
ListedSeptember 2026
WebVisit
Report a problem
Around the Web
Show HN2 source points · 0 comments