RecSoup: Merging pruned objective-specific recommenders to tune trade-offs without retraining
_reachsumit · x · 2026-10-16
The paper proposes RecSoup, an efficient model merging approach for controllable multi-objective recommendation. It first prunes conflicting parameters across objective-specific recommenders, then merges them, so trade-offs among objectives can be tuned at inference time without any retraining.
More from Research
- StepSafe: step-level guardrails teach agent monitors when to intervene — armancohan · 2026-10-17
- New algorithm cuts (min,+) matrix multiplication exponent to 2.9964, beating Anthropic's breakthrough — thegautamkamath · 2026-10-17
- 'Models get harder to understand with scale, not easier': new interp debate thread — NeelNanda5 · 2026-10-17
- What if LLMs could decompose trained models into explicit ordinary code? — tszzl · 2026-10-17
- ByteDance Seed Proposes Data-Space Iteration, a New Few-Step Generation Paradigm Beating Distillation Baselines — _akhaliq · 2026-10-17
- OmniCompass: 820-Video Benchmark Evaluates 50 MLLMs on Omni-Modal Video Understanding — _akhaliq · 2026-10-17