Research map / Optimization / Optimizers
SCRAPL: Scattering Transform with Random Paths for Machine Learning
Christopher Mitcheltree, Vincent Lostanlen, Emmanouil Benetos, Mathieu Lagrange
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 |
|---|---|---|---|
| add P-Adam path-wise moments (ablation) | SCRAPL · Granular synth | theta_synth L1 (per mille): 99.691 -> 87.433 | Improved |
| add P-SAGA (ablation) | SCRAPL + P-Adam · Granular synth | theta_synth L1 (per mille): 87.433 -> 73.809 | Improved |
| add theta-importance sampling (ablation) | SCRAPL + P-Adam + P-SAGA · Granular synth | theta_synth L1 (per mille): 73.809 -> 65.68 | Improved |
| add theta-importance sampling (ablation) | SCRAPL · Granular synth | convergence steps: 7296 -> 6014 | Improved |
| add theta-importance sampling (ablation) | SCRAPL · Granular synth theta_density | theta_density L1 (per mille): 70.422 -> 72.621 | Hurt |
| use theta-IS instead of uniform path sampling (ablation) | SCRAPL · Chirplet synth, AM 1-2 Hz, FM 2-4 oct/s | theta_AM L1 (per mille): 111.03 -> 55.474 | Improved |
| use theta-IS instead of uniform path sampling (ablation) | SCRAPL · Chirplet synth, AM 2.8-8.4 Hz, FM 2-4 oct/s | theta_FM L1 (per mille): 238.43 -> 48.464 | Improved |
| replace full JTFS with SCRAPL random-path loss | JTFS loss · Granular synth | theta_synth L1 (per mille): 42.4 -> 65.68 | Hurt |
| use SCRAPL instead of multiscale spectral losses | MSS losses · Granular synth | theta_synth L1 (per mille): 195.47 -> 65.68 | Improved |
| use SCRAPL instead of full JTFS | JTFS loss · Granular synth | computational cost | Improved |
| use SCRAPL (with theta-IS) in TR-808 task instead of MSS | MSS Linear · TR-808 micro, JTFS distance | JTFS distance: 778.73 -> 1054.2 | Hurt |
| use SCRAPL instead of MSS for misaligned audio | MSS · TR-808 meso | FAD (EnCodec): 3.3305 -> 2.4193 | Improved |
Closest papers by meaning
- Massive Sound Embedding Benchmark (MSEB) (NeurIPS 2025 · similarity 0.82)
- MDSGen: Fast and Efficient Masked Diffusion Temporal-Aware Transformers for Open-Domain Sound Generation (ICLR 2025 · similarity 0.82)
- Improving Sound Source Localization with Joint Slot Attention on Image and Audio (CVPR 2025 · similarity 0.82)
- RFWave: Multi-band Rectified Flow for Audio Waveform Reconstruction (ICLR 2025 · similarity 0.81)
- DiffSED: Sound Event Detection with Denoising Diffusion (AAAI 2024 · similarity 0.81)
- Multi-band Frequency Reconstruction for Neural Psychoacoustic Coding (ICML 2025 · similarity 0.81)
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.
