Research map / Label-efficient and robust learning / Self-supervised and contrastive
SURF: Separation via Unsupervised Remixing Flow
Henry Li, Robin Scheibler, Efthymios Tzinis, Matt Shannon, Arnaud Doucet, John R. Hershey
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Replace regression student with flow matching student in remixing | Self-Remixing · CIFAR10 (2 sources) | FID: 29.12 -> 14.77 | Improved |
| Replace regression student with flow matching student in remixing | ReMixIT · MNIST (2 sources) | PSNR: 22.81 -> 37.26 | Improved |
| Replace regression student with flow matching student in remixing | ReMixIT · Libri2Mix | SI-SDR: 13.22 -> 16.54 | Improved |
| Use SURF instead of BASIS prior-based separation | SURF (Self-Remixing) · CIFAR10 | FID: 22.12 -> 12.50 | Improved |
| Use SURF instead of BASIS | SURF (Self-Remixing) · CIFAR10 | Inception Score: 8.29 vs 8.25 | No clear effect |
| Use SURF instead of supervised flow | SURF (ReMixIT) · MNIST | PSNR: 37.44 vs 37.26 | No clear effect |
| Use SURF instead of supervised flow | SURF (ReMixIT) · CIFAR10 | FID: 9.601 vs 14.83 | Hurt |
| Apply SURF to AudioSet-trained model | SURF (ReMixIT) · FUSS 1-source | SI-SDR: 19.83 -> 32.67 | Improved |
| Apply SURF to AudioSet-trained model | SURF (ReMixIT) · FUSS 4-source | SI-SDR: 9.75 vs 7.52 (MixIT) | Hurt |
| Apply SURF to AudioSet-trained model | SURF · FUSS source counting | Equal: 0.387 -> 0.574 (ReMixIT vs SURF ReMixIT) | Improved |
Closest papers by meaning
- One Step Diffusion via Shortcut Models (ICLR 2025 · similarity 0.84)
- ZeroSep: Separate Anything in Audio with Zero Training (NeurIPS 2025 · similarity 0.83)
- Shortcutting Pre-trained Flow Matching Diffusion Models is Almost Free Lunch (NeurIPS 2025 · similarity 0.83)
- Faster Inference of Flow-Based Generative Models via Improved Data-Noise Coupling (ICLR 2025 · similarity 0.83)
- Inverse Flow and Consistency Models (ICML 2025 · similarity 0.83)
- Remix-DiT: Mixing Diffusion Transformers for Multi-Expert Denoising (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
