phenomenon · first seen 29 Oct 2026
neural thickets
The hypothesis that pretrained models sit in dense neighborhoods of task-specific experts reachable by random weight perturbation.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Rethinking Neural Thickets Without Weight PerturbationsPapers · 29 Oct 2026 · Nicolas Dufour, Elouan Gardès, Amil Dravid, Patrick Pérez, Loic Landrieu
- Does Random Search Discover Task Experts? Revisiting Neural ThicketsPapers · 31 Oct 2026 · Julian Asilis, Guanzhe Hong, Vatsal Sharan
- When Are Nearby Experts Composable? Cross-Task Model Merging in Neural ThicketsPapers · 8 Nov 2026 · Suxin Ji, Hungtao Wan, Mingjun Liu, An Zhang
Latest sources
- When Are Nearby Experts Composable? Cross-Task Model Merging in Neural ThicketsPapers · 8 Nov 2026 · Suxin Ji, Hungtao Wan, Mingjun Liu, An Zhang
- Does Random Search Discover Task Experts? Revisiting Neural ThicketsPapers · 31 Oct 2026 · Julian Asilis, Guanzhe Hong, Vatsal Sharan
- Rethinking Neural Thickets Without Weight PerturbationsPapers · 29 Oct 2026 · Nicolas Dufour, Elouan Gardès, Amil Dravid, Patrick Pérez, Loic Landrieu