Research map / Adversarial robustness and security / Certified and empirical robustness
Improved Certified Defenses against Data Poisoning with (Deterministic) Finite Aggregation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Reuse samples across base classifiers with Finite Aggregation (d=32) rather than disjoint DPA partitions | Network-In-Network · MNIST, d_sim ≤ 300 | certified fraction: $77.12\%$ -> $80.17\%$ | Helped |
| Reuse samples across base classifiers with Finite Aggregation (d=32) rather than disjoint DPA partitions | Network-In-Network · CIFAR-10, k=50, d_sim ≤ 10 | certified fraction: $46.44\%$ -> $50.31\%$ | Helped |
| Reuse samples across base classifiers with Finite Aggregation (d=32) rather than disjoint DPA partitions | Network-In-Network · GTSRB, k=100, d_sim ≤ 15 | certified fraction: $30.19\%$ -> $34.96\%$ | Helped |
| Increase spreading degree from d=1 to d=3 (ablation) | Finite Aggregation · CIFAR-10, k=250 | certified fraction: $-0.05\%$, $+0.07\%$, $-0.30\%$, $-0.10\%$, $+0.21\%$ | Mixed |
| Increase spreading degree from d=1 to d=8 (ablation) | Finite Aggregation · GTSRB, k=50 | certified fraction: $-0.09\%$, $-0.02\%$, $+0.09\%$, $-0.81\%$, $+2.56\%$ | Mixed |
| Increase spreading degree d (ablation) | Finite Aggregation · MNIST, CIFAR-10, and GTSRB | certified fraction: up to $3.05\%$, $3.87\%$, and $4.77\%$ | Helped |
| Increase spreading degree d (ablation) | Finite Aggregation · MNIST, CIFAR-10, and GTSRB | clean accuracy | No clear effect |
| Increase partition parameter k (ablation) | Finite Aggregation · MNIST, CIFAR-10, and GTSRB | clean accuracy | Hurt |
Closest papers by meaning
- Deep Partition Aggregation: Provable Defenses against General Poisoning Attacks (ICLR 2021 · similarity 0.93)
- Lethal Dose Conjecture on Data Poisoning (NeurIPS 2022 · similarity 0.92)
- Run-off Election: Improved Provable Defense against Data Poisoning Attacks (ICML 2023 · similarity 0.91)
- Defending Against Sophisticated Poisoning Attacks with RL-based Aggregation in Federated Learning (AAAI 2025 · similarity 0.87)
- Dual Defense: Enhancing Privacy and Mitigating Poisoning Attacks in Federated Learning (NeurIPS 2024 · similarity 0.87)
- RFLPA: A Robust Federated Learning Framework against Poisoning Attacks with Secure Aggregation (NeurIPS 2024 · similarity 0.85)
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.