srush derives MNIST in 3,000 lines of Lean, minimal DL math is half a semester
srush_nlp · x · 2026-10-29
Sasha Rush (srush) released 'Deep Learning Speedrun', a Lean formalization deriving MNIST training from first principles, answering: what's the minimal math needed for deep learning?
- Answer: about half of one freshman semester — Harvard Math 55 with everything useless removed (integrals included); no linear algebra required.
- Roughly 3,000 lines of Lean, aligned to 'Baby Rudin', building from metric spaces and topology down to 'move towards the downwards direction'.
- Framed against Terence Tao's quip that any undergrad math major has the prerequisites for LLM math.
- Author self-deprecatingly labels the blog 'AI slop' and challenges others to derive it faster.
Related event: srush Derives MNIST from Peano Axioms in 3000 Lines of Lean(2 posts)→
More from Research
- pstack's /statechart skill ships with a mini TLC compiler that helped find real distributed-systems bugs — DavidKPiano · 2026-10-30
- Neural twins: models trained for each other, chatting in neuralese — repligate · 2026-10-30
- Self-Direction turns agent mistakes into LoRA lessons, lifting GAIA success from 38.8% to 55.8% — rohanpaul_ai · 2026-10-30
- Preprint: AlphaGenome pervasively underestimates causal variant effects, undermining fine-mapping — anshulkundaje · 2026-10-30
- One Researcher on 22 NeurIPS 2026 Papers: What Should Author Lists Mean? — qberthet · 2026-10-30
- Stanford's I3L boosts whole-body robot deployment success by 68.3 points — jiajunwu_cs · 2026-10-30