FD-Loss Achieves 0.75 One-Step Pixel-Space FID, Accepted as NeurIPS Oral
yuewang314 · x · 2026-09-25
PhD student Jiawei Yang announced FD-Loss has been accepted as an Oral at NeurIPS, capping off his PhD. The method directly optimizes the FID metric: the 0.9 FID one-step pixel-space result he cryptically teased two months ago has now improved to 0.75, with room to go lower. The research started from a simple question—can FID be optimized directly, and what does that reveal—beginning as what he calls a deliberate small FID prank.
More from Research
- GRAM accepted to NeurIPS 2026: recursive reasoning becomes stochastic latent trajectories — SungjinAhn_ · 2026-09-25
- Photonic matrix core on thin-film lithium niobate runs in-situ backpropagation at 8-bit precision — jwt0625 · 2026-09-25
- Agentic robotics keeps bolting GPT-6 onto harnesses — but where are systems that learn from physical failure? — DJiafei · 2026-09-25
- Dev Proposes "Evidentiality Framework": Provenance Markers for Every LLM Claim — jzesbaugh · 2026-09-25
- New Paper Quantifies When Rubric-Based RL Genuinely Improves Models vs. Hacks the Verifier — heghbalz · 2026-09-25
- DeepSeek's DSec sandbox platform serves 3M sandboxes daily for agent RL training; Liang Wenfeng co-authors — teortaxesTex · 2026-09-25