Research map / Learning theory and algorithms / Combinatorial and search optimization
A Unified Framework for Data Poisoning Attack to Graph-based Semi-supervised Learning
Xuanqing Liu, Si Si, Xiaojin Zhu, Yang Li, Cho-Jui Hsieh
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 |
|---|---|---|---|
| Apply a carefully designed label perturbation with ||δy||=3 | Label propagation · cadata | RMSE: 0.2 to 0.3 | Hurt |
| Flip 3 labeled nodes using the proposed attack | Label propagation · mnist17 | accuracy: 98.46% to 50% | Hurt |
| Flip 2 labeled data | Graph-based semi-supervised learning · MNIST binary classification | error rate: around 50% | Hurt |
| Use estimated unlabeled labels instead of ground-truth labels | Proposed label-poisoning attack · cadata and E2006 | RMSE | No clear effect |
| Use estimated unlabeled labels instead of ground-truth labels | Proposed label-poisoning attack · mnist17 and rcv1 | error rate | No clear effect |
| Use the probabilistic solver instead of greedy search | Classification poisoning attack · rcv1 | error rate | Improved |
| Increase the number of labeled data | Label propagation under poisoning · all four datasets | robustness to poisoning | Improved |
| Use an imperfect estimate of γ, with γadv=1/8 γtruth | Proposed label-poisoning attack · cadata | RMSE: 0.223 to 0.218 | Hurt |
| Enforce an ℓ0 sparsity constraint on the label perturbation | Sparse PCA poisoning solution · cadata and E2006 | RMSE | Mixed |
| Use the poisoning attack in transductive versus inductive evaluation | Manifold regularization · cadata | RMSE | No clear effect |
Closest papers by meaning
- The Perils of Learning From Unlabeled Data: Backdoor Attacks on Semi-supervised Learning (ICCV 2023 · similarity 0.88)
- Lethal Dose Conjecture on Data Poisoning (NeurIPS 2022 · similarity 0.87)
- Not All Poisons are Created Equal: Robust Training against Data Poisoning (ICML 2022 · similarity 0.87)
- Deterministic Certification of Graph Neural Networks against Graph Poisoning Attacks with Arbitrary Perturbations (CVPR 2025 · similarity 0.85)
- Clean-Label Graph Backdoor Attack in the Node Classification Task (AAAI 2025 · similarity 0.85)
- Activation Gradient based Poisoned Sample Detection Against Backdoor Attacks (ICLR 2025 · 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.
