Research question
Can outsourced training insert hidden behaviour into a network?
Key findings
- Trigger patterns cause targeted misclassification while clean accuracy stays high - Backdoors survive transfer learning to new tasks
BadNets
Introduced neural network backdoors and framed them as a supply-chain problem in outsourced and pre-trained models, later published in IEEE Access.
Contribute
Spot an error, or know a record we should add? Members can propose additions and submit corrections.