Two NeurIPS 2026 papers: margin conditions tame sequential task confusion
khademinori · x · 2026-09-25
Two papers accepted at NeurIPS 2026. The first proves that under a Mahalanobis margin condition on pretrained representations, task confusion between classes learned at different times decays exponentially in the margin, and that fine-tuning erodes the margin at a rate growing with the learning rate. The second gives a closed-form characterization of the per-stage offset L2 regularization induces in sequentially trained cross-entropy heads, introducing gauge anchoring.
More from Research
- TANGO from UC Berkeley and Princeton brings whole-body AI control to humanoid robot navigation — shahdhruv_ · 2026-09-25
- Visualizing Raw Activation Values Could Seed a Math Framework for Interpretability — brandon_xyzw · 2026-09-25
- NVIDIA Talks Cosmos 3: Simulation and Verifiable Rewards for Physical AI — liu_mingyu · 2026-09-25
- Open-source fly Matrix: 150,802-neuron connectome flies in your browser — mtizard · 2026-09-25
- IDS paper wins NeurIPS oral: AI writes formally verified distributed systems 200x faster — AccBalanced · 2026-09-25
- NeurIPS desk-rejects 5/5/5-scored paper over one co-author's policy violation — AccBalanced · 2026-09-25