AI Paper Atlas

Research map / Optimization / Optimizers

SCRAPL: Scattering Transform with Random Paths for Machine Learning

Christopher Mitcheltree, Vincent Lostanlen, Emmanouil Benetos, Mathieu Lagrange

ICLR 2026 · Optimization › Optimizers · approach: Surrogate losses and robust solvers

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 changedWhereMeasuredEffect
add P-Adam path-wise moments (ablation)SCRAPL · Granular synththeta_synth L1 (per mille): 99.691 -> 87.433Improved
add P-SAGA (ablation)SCRAPL + P-Adam · Granular synththeta_synth L1 (per mille): 87.433 -> 73.809Improved
add theta-importance sampling (ablation)SCRAPL + P-Adam + P-SAGA · Granular synththeta_synth L1 (per mille): 73.809 -> 65.68Improved
add theta-importance sampling (ablation)SCRAPL · Granular synthconvergence steps: 7296 -> 6014Improved
add theta-importance sampling (ablation)SCRAPL · Granular synth theta_densitytheta_density L1 (per mille): 70.422 -> 72.621Hurt
use theta-IS instead of uniform path sampling (ablation)SCRAPL · Chirplet synth, AM 1-2 Hz, FM 2-4 oct/stheta_AM L1 (per mille): 111.03 -> 55.474Improved
use theta-IS instead of uniform path sampling (ablation)SCRAPL · Chirplet synth, AM 2.8-8.4 Hz, FM 2-4 oct/stheta_FM L1 (per mille): 238.43 -> 48.464Improved
replace full JTFS with SCRAPL random-path lossJTFS loss · Granular synththeta_synth L1 (per mille): 42.4 -> 65.68Hurt
use SCRAPL instead of multiscale spectral lossesMSS losses · Granular synththeta_synth L1 (per mille): 195.47 -> 65.68Improved
use SCRAPL instead of full JTFSJTFS loss · Granular synthcomputational costImproved
use SCRAPL (with theta-IS) in TR-808 task instead of MSSMSS Linear · TR-808 micro, JTFS distanceJTFS distance: 778.73 -> 1054.2Hurt
use SCRAPL instead of MSS for misaligned audioMSS · TR-808 mesoFAD (EnCodec): 3.3305 -> 2.4193Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy