Research map / Label-efficient and robust learning / Recognition and long-tail
Model Ratatouille: Recycling Diverse Models for Out-of-Distribution Generalization
Alexandre Rame, Kartik Ahuja, Jianyu Zhang, Matthieu Cord, Leon Bottou, David Lopez-Paz
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 |
|---|---|---|---|
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · PACS | accuracy: 88.7 ± 0.2 -> 89.5 ± 0.1 | Improved |
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · VLCS | accuracy: 79.1 ± 0.1 -> 78.5 ± 0.1 | Hurt |
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · OfficeHome | accuracy: 72.1 ± 0.2 -> 73.1 ± 0.1 | Improved |
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · TerraInc | accuracy: 51.4 ± 0.6 -> 51.8 ± 0.4 | Improved |
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · DomainNet | accuracy: 47.4 ± 0.2 -> 47.5 ± 0.1 | No clear effect |
| Recycle diverse auxiliary fine-tunings as separate target-task initializations, then average the target-fine-tuned weights | Model soups · DomainBed Avg | accuracy: 67.6 -> 68.1 | Improved |
| Use greedy rather than uniform weight selection (ablation) | Model ratatouille · TerraInc | accuracy: 51.8 ± 0.4 -> 49.2 ± 0.9 | Hurt |
| Increase the number of auxiliary tasks (ablation) | Model ratatouille · DomainBed OOD test domains | accuracy | Improved |
| Use auxiliary-task recycling instead of extending target fine-tuning alone (ablation) | Model ratatouille vs. model soups · OfficeHome, Art OOD domain | accuracy | Improved |
| Increase the number of target fine-tuning runs (ablation) | Model ratatouille · DomainBed OOD test domains | accuracy | Improved |
| Use moving-average checkpoints for auxiliary-task initializations (ablation) | Model ratatouille · DomainBed Avg | accuracy: 68.1 -> 68.2 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- SWAD — SWAD: Domain Generalization by Seeking Flat Minima (NeurIPS 2021) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Understanding the Effects of RLHF on LLM Generalisation and Diversity (ICLR 2024 · similarity 0.85)
- MixEval-X: Any-to-any Evaluations from Real-world Data Mixture (ICLR 2025 · similarity 0.85)
- SKILL-MIX: a Flexible and Expandable Family of Evaluations for AI Models (ICLR 2024 · similarity 0.85)
- Prismatic Synthesis: Gradient-based Data Diversification Boosts Generalization in LLM Reasoning (NeurIPS 2025 · similarity 0.84)
- Towards Few-Shot Adaptation of Foundation Models via Multitask Finetuning (ICLR 2024 · similarity 0.84)
- ThetaEvolve: Test-time Learning on Open Problems (ICML 2026 · similarity 0.84)
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.
