Research map / Federated learning and privacy / Privacy attacks and secure learning
Incentives in Private Collaborative Machine Learning
Rachael Hwee Ling Sim, Yehong Zhang, Trong Nghia Hoang, Xinyi Xu, Bryan Kian Hsiang Low, Patrick Jaillet
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 |
|---|---|---|---|
| weaken party 2's DP guarantee (increase epsilon_2 from 0.004 to 12.5) | Bayesian surprise valuation with rho-Shapley reward · Syn, CaliH, Diab | party 2 valuation v_2, Shapley value, attained reward r_2 | Improved |
| strengthen party 2's DP guarantee (decrease epsilon_2) | grand coalition and party 2 individual models · Syn, CaliH, Diab | MNLP (lower is better) | Hurt |
| enforce privacy-valuation/reward trade-off: MNLP_r of reward rises as epsilon_2 decreases | proposed scheme · Syn, CaliH, Diab | MNLP_r of party 2's model reward | Hurt |
| join collaboration (reward model) instead of training alone, as epsilon_2 decreases | party 2 · Syn, CaliH, Diab | MNLP_r vs MNLP_i | Improved |
| exception: privacy-utility monotonicity under validation-based valuation | validation-set-based valuation · Diab | MNLP vs epsilon_2 | No clear effect |
| replace noise addition (tau_i) with likelihood tempering (kappa_i) for reward control (ablation) | reward control mechanism · Syn | monotonicity of r_i | Improved |
| replace noise addition with likelihood tempering (ablation) | reward control mechanism · Syn | similarity r'_i to grand coalition posterior at same r_i | Improved |
| replace noise addition with likelihood tempering (ablation) | reward control mechanism · Syn, CaliH, Diab | MNLP_r (lower is better) | Improved |
| replace noise addition with likelihood tempering (ablation) | reward control mechanism · Syn, CaliH, Diab | variance of MNLP_r | Improved |
| other parties choose weaker privacy (epsilon = 2 instead of 0.2) | party 2 utility gain · appendix experiments | party 2 utility gain | Improved |
| use non-noise-aware inference instead of DP noise-aware inference (ablation) | valuation function · appendix | privacy-valuation trade-off | Hurt |
Closest papers by meaning
- Collaborative Machine Learning with Incentive-Aware Model Rewards (ICML 2020 · similarity 0.89)
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.89)
- Distributed Learning without Distress: Privacy-Preserving Empirical Risk Minimization (NeurIPS 2018 · similarity 0.88)
- It's My Data Too: Private ML for Datasets with Multi-User Training Examples (ICML 2025 · similarity 0.88)
- Incentivizing Time-Aware Fairness in Data Sharing (NeurIPS 2025 · similarity 0.88)
- Paid with Models: Optimal Contract Design for Collaborative Machine Learning (AAAI 2025 · similarity 0.87)
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.
