Research map / Adversarial robustness and security / Attacks on text and graphs
Tools for Verifying Neural Models' Training Data
Dami Choi, Yonadav G Shavit, David Duvenaud
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 |
|---|---|---|---|
| use memorization test with 1% of data to detect glued runs | memorization plot · GPT-2 gluing attack | long-term memorization | Improved |
| use memorization test on interpolated checkpoints | memorization plot · GPT-2 interpolation attack, 1% data | memorization diagonal | Improved |
| use memorization test to detect data addition | memorization test · GPT-2 data addition, 1% samples | memorization | No clear effect |
| detect data addition by retraining suspicious segments | segment retraining · GPT-2 data addition (50% same distribution or 5% Github) | weight distance after retraining | Improved |
| use subtraction upper bound lambda with 1% of data | lambda test · GPT-2 data subtraction 0%, 5%, 50%, 95% | lambda upper bound | Improved |
| increase model size from Pythia 70M to 1B | memorization effect · Pythia | memorization clarity | Improved |
| reduce data sampling rate to 1% or 0.1% (ablation) | memorization plot · GPT-2 and Pythia | memorization effect visibility: 1% and 0.1% | Mixed |
| later training checkpoints in Pythia 70M | memorization plot · Pythia 70M with low sampling | memorization effect visibility | Hurt |
Closest papers by meaning
- Training Verification-Friendly Neural Networks via Neuron Behavior Consistency (AAAI 2025 · similarity 0.87)
- Certified but Private: Scalable Zero-Knowledge Proofs for the Formal Verification of Neural Networks (NeurIPS 2026 · similarity 0.86)
- Training Data Provenance Verification: Did Your Model Use Synthetic Data from My Generative Model for Training? (CVPR 2025 · similarity 0.85)
- VNN: Verification-Friendly Neural Networks with Hard Robustness Guarantees (ICML 2024 · similarity 0.85)
- Efficiently Verifiable Proofs of Data Attribution (NeurIPS 2025 · similarity 0.84)
- Verification of Machine Unlearning is Fragile (ICML 2024 · similarity 0.84)
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.
