Research question
Can an attacker degrade a classifier by injecting crafted training points?
Key findings
- Gradient-ascent crafted points substantially increase SVM test error - Established the optimisation formulation of poisoning
Poisoning Attacks against Support Vector Machines
The first gradient-based poisoning attack, introducing the optimisation view of training-data attacks that later work on deep networks and LLMs builds on.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.