Paper recasts evolution-strategy updates as center-of-mass shifts
The mean-subtracted update matches a reward-weighted displacement for any finite population when rewards are nonnegative.
Stories per day · 30 days
5 RSS
Relatedreinforcement-learning2evolution-strategies1open-ended-generation1optimization1rubric-rewards1
10 stories, newest first · one card per story
Stories tagged #training-methods
HarnessRL co-evolves open-ended RL rewards with the policy
On Rubric-Mix-10k it beats the strongest baselines by 2.8 to 3.1 points and nearly halves steps to the best checkpoint.
Calibrating Teacher--Student Discrepancy for On-Policy Distillation
Researchers propose Cal-OPD, a calibrated on-policy distillation method that filters teacher self-deviation and improves reasoning model training.
RetireOPD: Self-Retiring On-Policy Distillation for Agentic Reinforcement Learning
RetireOPD trains multi-turn agents via self on-policy distillation with adaptive teacher retirement, lifting Qwen2.5 ALFWorld success by up to 18.8%.
Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL
ActObs supervises observation tokens during SFT, improving GRPO-trained agent pass@k on Terminal-Bench 2.0 and cross-domain code editing for Qwen3 models.
Coupled Calibration and Learning: Mitigating Teacher Bias in LLM Distillation without Target-Domain Reward Feedback
CCL couples teacher calibration with student updates via token-level branching, provably removing teacher bias in LLM distillation.
Sakana AI Researchers Introduce PC-ALM, a Layer-Local Alternative to Backpropagation That Trains 1000-Layer Networks
Sakana AI's PC-ALM adds per-layer Lagrange multipliers to predictive coding, matching backprop on networks up to 1000 layers with layer-local updates.
6
Negative Self-Distillation: Learning to Reason by Avoiding Flaws
Researchers propose Negative Self-Distillation (NSD), a label-free LLM self-improvement method that diverges from self-generated flawed reasoning rather than imitating privileged solutions.
Kalman Delta Networks: Uncertainty-aware Associative Memory
Researchers propose Kalman Delta Networks, adding Kalman-filter uncertainty tracking to delta-rule linear attention, improving perplexity and accuracy at 750M and 1.3B parameters.
Granite 4.2 LLMs: How They're Built
IBM releases Granite 4.2 LLMs with a Hugging Face post detailing how the model family was built.