Energy-Guided Flow Matching Cuts Training Cost, Improves Quality
Haoyang Tong · hf · 2026-08-19
Introduces Energy-Guided Flow Matching, a method that improves generative quality by progressively revealing high-frequency details through a moving endpoint and adaptive scheduling. It reduces training costs and achieves state-of-the-art FID scores.
More from Research
- Rényi entropies uniquely determine probability distributions — abeirami · 2026-08-19
- Paper reassesses Muon optimizer for large-scale matrix factorization — YouJiacheng · 2026-08-19
- Personalized Auto-Research Framework Towards True AI Co-Scientist — Bo Ni · 2026-08-19
- Embodied-Navigator: Point, Think, Memorize, and Align for Navigation — OmniAI-ZJU · 2026-08-19
- Paper Unifies GNN Architectures Through Common Layer Equation — Sai Karthik Navuluru · 2026-08-19
- Paper proposes stable transport-mechanism descriptor for per-pixel rendering difficulty — ssh4net · 2026-08-19