AI Paper Atlas

Research map / Large language models / Alignment and preferences

MARS-Sep: Multimodal-Aligned Reinforced Sound Separation

Zihan Zhang, Xize Cheng, Zhennan Jiang, Dongjie Fu, Jingyuan Chen, Zhou Zhao, Tao Jin

ICLR 2026 · Large language models › Alignment and preferences · approach: Steering and correcting LLM behavior

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 RL with multimodal reward on top of supervised OmniSep (frozen encoder) (ablation)OmniSep · VGGSound-clean+ textSDR: 6.70 -> 6.71Improved
add progressive fine-tuning of encoder with supervised loss only (ablation)OmniSep · VGGSound-clean+ textSDR: 6.70 -> 0.75Hurt
add RL to the fine-tuned encoder (full model) (ablation)MARS-Sep · VGGSound-clean+ textSDR: 0.75 -> 6.91Improved
use 3-stage instead of no fine-tuning of the reward encoder (ablation)MARS-Sep · VGGSound-clean+ audio querySDR: 6.85 -> 7.33Improved
use 3-stage instead of 1-stage fine-tuning (ablation)MARS-Sep · VGGSound-clean+ image querySDR: 6.54 -> 6.93Mixed
use MLBP query pooling instead of max pooling (ablation)MARS-Sep · MUSIC-clean+ textCLAP_t: 5.34 -> 6.18Improved
use MLBP instead of learned weighted sums (ablation)MARS-Sep · VGGSound-clean+ composed querySDR: 8.08 vs 7.93Mixed
increase Beta concentration kappa from 3 to 9 (ablation)MARS-Sep · VGGSound-clean+ composed querySDR: 6.99 -> 7.93Improved
vary PPO clip range epsilon 0.01-0.4 (ablation)MARS-Sep · VGGSound-clean+ composed querySDR: 7.90 to 7.93No clear effect
raise entropy coefficient from 0.01 to 0.2 (ablation)MARS-Sep · VGGSound-clean+ composed querySDR: 7.20 -> 7.93Improved
add KL penalty 0.1 vs none (ablation)MARS-Sep · VGGSound-clean+ composed querySDR: 7.43 -> 7.93Mixed
replace OmniSep supervised mask regression with MARS-SepOmniSep · MUSIC-clean+ audio querySDR: 10.37 -> 11.73Improved

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