Training ideas

Curricula, sampling schedules, pseudo-labeling and multi-task mixing, with the standard training recipe as the control.

2 ideas found

Status: Beats tuned baseline Category: Training ML: Training dynamics Clear all

✓✓ Beats tuned baseline 2026

Phase-Aware Bias-Energy Trust Region

Use the envelope's phase transition to choose whether clipping should primarily control update energy or preserve the raw gradient and reduce clipping bias. In the energy-dominated regime, regulate the retained update energy; in the bias-dominated regime, regulate the removed-gradient residual and monitor rare outliers explicitly.

Useful7/10

Difficulty5/10

Novelty5/10

Training Stability Accuracy Generalization

Paper: A Sharp Joint Bias-Energy Envelope for Radial Clipping arXiv:2608.18999

✓✓ Beats tuned baseline 2026

Spectral-Width Prethermal Training Schedule

Use the paper's lifetime law as a controller for training or rollout difficulty. Estimate the active perturbation bandwidth R of hidden states or forecast errors and reduce the residual gain, increase the dispersion order W, or inject controlled bandwidth whenever the estimated prethermal lifetime becomes too short.

Useful7/10

Difficulty5/10

Novelty8/10

Training Stability Speedup Generalization

Paper: From stable periodic orbits to many-body chaos: doubly tunable prethermalization via engineering of an emergent band structure arXiv:2607.12355