Sakana AI Paper: Neural Networks Can Learn While Respecting Biological Principles
simonguozirui · x · 2026-07-31
Explores whether building AI requires exactly copying the brain's structure. The human brain follows Dale's principle, where each neuron is strictly either excitatory or inhibitory, a rule most artificial neural networks ignore.
A paper by Sakana AI investigated if neural networks can learn difficult tasks while strictly obeying this rule:
- Researchers separated excitatory and inhibitory signals and abandoned standard backpropagation.
- They used an Error Diffusion (ED) method that doesn't require transporting weights backward.
- The network achieved 96.7% accuracy on MNIST and 61.7% on CIFAR-10.
This shows biologically plausible learning is possible, though not yet matching conventional networks. Like how building airplanes didn't require mimicking bird wings, studying the brain can still offer valuable inspirations for AI.
Related event: Sakana AI Explores Biologically-Constrained Neural Networks(3 posts)→
More from Research
- Goodfire's Nick Cammarata argues InceptionV1 polysemanticity wasn't caused by superposition — petrusenko_max · 2026-10-16
- UPenn lab shows 'critique and revise' prompting doubles Claude's research ideation output — arjunrajlab · 2026-10-16
- LLM ideation experiment: just ask for non-obvious ideas — and Opus beats smaller models by a wide margin — arjunrajlab · 2026-10-16
- What prompt gets the most interesting ideas out of LLMs? A researcher's ongoing experiment — arjunrajlab · 2026-10-16
- SimpleMark: multi-bit LLM watermark bounds per-token distortion with f-divergence budget — rishabh16_ · 2026-10-16
- Stanford paper: generative representation learning cuts tokenizer training compute by 7-8x — HirokatuKataoka · 2026-10-16