ExploreNet's gain comes from targeted exploration, not larger noise, ablation shows
StellaLisy · x · 2026-10-08
ExploreNet's learned noise averages 1.4x the magnitude of FlowGRPO's isotropic noise. Matching magnitude but making the noise isotropic removes the gain, proving targeted exploration — not noise size — drives the improvement. ExploreNet also works with smaller group sizes and fewer denoising steps.
More from Research
- Tencent's STEPQuant: 6-bit recurrent states match FP32 with 68.7% less memory — _akhaliq · 2026-10-09
- Stanford's Level-of-Token Diffusion cuts image and video generation cost with multiresolution tokens — GordonWetzstein · 2026-10-09
- Datapoint launches Streams: live human feedback RL for multimodal models — _akhaliq · 2026-10-09
- Agent Plasticity: top-performing agents aren't the most efficient learners — RulinShao · 2026-10-09
- BiGym 2.0: LLM-Written Policies Hit 65% From One Demo, But π0.5 Still Leads at 75% — stepjamUK · 2026-10-09
- Jonathon Stray seeks engineer for large-scale study of long-term AI use on young adults — RishiBommasani · 2026-10-09