Follow-up to NeurIPS Oral: derivative-based losses improve Strong Stochastic Flow Maps by 10-50%
AlexanderTong7 · x · 2026-10-22
A follow-up to the NeurIPS 2026 Oral "Improved Training of Strong Stochastic Flow Maps": the team improves SSFM training by unlocking derivative-based losses.
- They explore the loss design space for SSFMs using a random ODE approach.
- This improves FD Dino by 10-50% over the original implementation.
- Joint work across multiple institutions.
More from Research
- ALPS: a single visual encoder unifying state, language, and pixels for world models — cihangxie · 2026-10-23
- EmbeddingGemma 2 used for speech-to-text: 1 hour of audio transcribed locally in ~13 seconds — GlennCameronjr · 2026-10-23
- Researchers eye open-model training data attribution as LLMs shown memorizing Cas9 and Alu sequences — anshulkundaje · 2026-10-23
- Motoniq's self-reflective VLA hits 92% on Perturbed LIBERO, beating RHO (53%) and Claude agent (9%) — hbouammar · 2026-10-23
- Sasha Rush mocks 'vibed ML papers' that bolt on a 'Theoretical Analysis' section — srush_nlp · 2026-10-23
- LSA: learnable spectral activations beat SIREN and FINER on INR reconstruction — orlitany · 2026-10-23