ICML keynote touts Weight Space Learning: 'Stable Diffusion for neural network weights'
burny_tech · x · 2026-11-05
At the ICML Workshop on Weight-Space Symmetries, keynote speaker Damian Borth argued that neural network weights are emerging as a novel modality in machine learning. His slides outline the research area of "weight-space programming": editing other networks' weights directly, which he likens to pixel-by-pixel image editing — with far more potential if done in a latent space of weights. BorisAKnyazev sums it up: Weight Space Learning is "Stable Diffusion for weights."
More from Research
- RoboColosseum founder: robotics lags AI because it lacks real-world benchmarks — DJiafei · 2026-11-05
- ByteDance study finds on-policy distillation leaves models 2.45-16.25% below base on out-of-distribution tasks — ByteDance · 2026-11-05
- FrontierPhysics: 56 expert tasks benchmark AI agents on end-to-end physics research — Bingran You · 2026-11-05
- VideoMemBench: new benchmark exposes major gaps in MLLMs' in-context video memory — Zhengdong Hu · 2026-11-05
- Muon-RT swaps momentum for teacher trajectory, cutting memory and beating AdamW by 2.6 points — IITK · 2026-11-05
- Self-Harness: Shanghai AI Lab's Method Lets Agents Fix Themselves, 30.2%→84.6% Without Retraining — Promptmethus · 2026-11-05