MLP paper shows neurons turn monosemantic in clustered regression, challenging global subspace view
burkov · x · 2026-09-07
This paper challenges the common view that neural networks learn a single global low-dimensional subspace of predictive directions. For clustered data — where each cluster has its own predictive direction and response function — local prediction can be simple while the directions collectively span the full ambient dimension, leaving no useful global low-dimensional structure.
Studying standard MLPs on this clustered regression setting, the authors show the networks develop a different form of feature learning: individual neurons become monosemantic, each strongly aligning with the predictive direction of one specific cluster. The network thereby discovers both the latent cluster structure and the per-cluster prediction rules.
More from Models
- OpenAI's Astra impresses with 3D interactions: directional sticker peeling demo — arena · 2026-09-08
- sanoTTS: a 337KB neural TTS that runs on a $3 ESP32-S3 or in your browser — alexcovo_eth · 2026-09-08
- Astra saturates Epoch AI's EBR-bench with 100% score in two playthroughs, beating top human — Jsevillamol · 2026-09-08
- Grok 4.7 reportedly being stealth-tested in Grok Bot ahead of release — mark_k · 2026-09-08
- Chess benchmarks questioned: a model could just download Stockfish and crush Magnus — iruletheworldmo · 2026-09-08
- Claude Fable 5.1 system prompt reveals why Claude 5 models wrote so badly — dbreunig · 2026-09-08