Research map / Learning theory and algorithms / Prediction and decision losses
On the Power of Source Screening for Learning Shared Feature Extractors
Muxing Wang, Connor Mclaughlin, Lili Su
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 source screening before FedRep training | Full Population training · ACSIncome (linear) | accuracy (%): 72.8 -> 74.2 | Improved |
| add source screening before FedRep training | Full Population training · CelebA smile | accuracy (%): 89.5 -> 90.5 | Improved |
| increase assumed dimension k from 3 to 7 in screening (ablation) | Ours · ACSIncome | accuracy (%): 73.4 -> 74.2 | Improved |
| increase assumed dimension k from 3 to 7 in screening (ablation) | Ours · CelebA | accuracy (%): 90.0 -> 90.5 | Improved |
| use random selection of sources instead of screening | Full Population training · ACSIncome | accuracy (%): 72.8 -> 71.2 | Hurt |
| use power-of-choice selection | Full Population training · ACSIncome | accuracy (%): 72.8 -> 73.0 | No clear effect |
| add source screening with two-layer NN extractor | Full Population training · ACSIncome (d_h=64) | mean accuracy (%): 79.91 -> 80.13 | Improved |
| add source screening with two-layer NN extractor | Full Population training · ACSIncome (d_h=256) | mean accuracy (%): 79.49 -> 79.96 | Improved |
| reweight clients by inverse diversity (V1) instead of subsampling (ablation) | Source subsampling · ACSIncome (d_h=256) | mean accuracy (%): 79.96 -> 79.38 | Hurt |
| reweight clients by inverse diversity and data volume (V2) instead of subsampling (ablation) | Source subsampling · ACSIncome (d_h=256) | mean accuracy (%): 79.96 -> 79.58 | Hurt |
| reweight clients (V1) versus full population (ablation) | Full Population training · ACSIncome (d_h=256) | std of accuracy: 0.06 -> 0.03 | Improved |
Closest papers by meaning
- Few-Shot Learning via Learning the Representation, Provably (ICLR 2021 · similarity 0.86)
- From Shared Subspaces to Shared Landmarks: A Robust Multi-Source Classification Approach (AAAI 2017 · similarity 0.86)
- Subspace Fitting Meets Regression: The Effects of Supervision and Orthonormality Constraints on Double Descent of Generalization Errors (ICML 2020 · similarity 0.85)
- Sample-Efficient Linear Representation Learning from Non-IID Non-Isotropic Data (ICLR 2024 · similarity 0.84)
- Joint‑Embedding vs Reconstruction: Provable Benefits of Latent Space Prediction for Self‑Supervised Learning (NeurIPS 2025 · similarity 0.83)
- Learning Shared Representations from Unpaired Data (NeurIPS 2025 · similarity 0.83)
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.
