Research map / Label-efficient and robust learning / Recognition and long-tail
Building a Winning Team: Selecting Source Model Ensembles using a Submodular Transferability Estimation Approach
Vimal K B, Saketh Bachu, Tanmay Garg, Niveditha Lakshmi Narasimhan, Raghavan Konuru, Vineeth N Balasubramanian
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 the cohesion term W_C to MS-LEEP (ablation) | MS-LEEP · Fully-supervised pool avg | Weighted Kendall tau: 0.275 -> 0.265 (WKT); 0.221 -> 0.246 (KT); 0.348 -> 0.383 (PCC) | Mixed |
| add the cohesion term W_C to E-LEEP (ablation) | E-LEEP · Fully-supervised pool avg | Weighted Kendall tau: 0.097 -> 0.301 | Improved |
| include the cohesion term W_C alongside domain and task terms (ablation) | OSBORN · Fully-supervised pool avg | Pearson: 0.496 -> 0.552 | Improved |
| include the domain term W_D alongside task and cohesion terms (ablation) | OSBORN · Fully-supervised pool avg | Pearson: 0.482 -> 0.552 | Improved |
| include the task term W_T alongside domain and cohesion terms (ablation) | OSBORN · Fully-supervised pool avg | Pearson: 0.556 -> 0.552 | No clear effect |
| weight W_D more heavily than W_T and W_C (ablation) | OSBORN · Caltech101 | correlation | Improved |
| use Frobenius norm regularizer in OT for self-supervised pool (ablation) | OSBORN · Self-supervised pool avg | Weighted Kendall tau: 0.328 -> 0.365 | Improved |
| use Frobenius norm regularizer in OT for fully-supervised pool (ablation) | OSBORN · Fully-supervised pool avg | Weighted Kendall tau: 0.540 -> 0.547 | No clear effect |
| replace MS-LEEP/E-LEEP with OSBORN for choosing the ensemble | ensemble selection · StanfordCars self-supervised | accuracy: 6.056 -> 61.820 | Improved |
| replace MS-LEEP/E-LEEP with OSBORN for choosing the ensemble | ensemble selection · Caltech101 fully-supervised | accuracy: 68.533 -> 68.533 | No clear effect |
| select ensemble greedily with submodular maximization instead of exhaustive search (ablation) | OSBORN · Fully-supervised pool (3 targets avg) | accuracy: 79.857 -> 76.315 | Mixed |
| use ensemble size larger than 3 (ablation) | ensemble selection · Oxford-IIIT Pets | test accuracy | No clear effect |
Closest papers by meaning
- Transferability Metrics for Selecting Source Model Ensembles (CVPR 2022 · similarity 0.94)
- How Stable Are Transferability Metrics Evaluations? (ECCV 2022 · similarity 0.88)
- H-ensemble: An Information Theoretic Approach to Reliable Few-Shot Multi-Source-Free Transfer (AAAI 2024 · similarity 0.87)
- Implicit Modeling for Transferability Estimation of Vision Foundation Models (NeurIPS 2025 · similarity 0.87)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.86)
- A High-Dimensional Statistical Method for Optimizing Transfer Quantities in Multi-Source Transfer Learning (NeurIPS 2025 · similarity 0.86)
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.
