Certifying Neural Network Robustness to Random Input Noisefrom Samples

Image credit: John Moeses Bauan

The authors proposed a certification method to upper bound the propability of misclassification when the input noise follows an arbitrary probability distribution. The problem is formulated as a chance-contrained optimization problem which can be solved via scenario approach.

Tianle Zhang
Tianle Zhang
PhD Student in Computer Science

My current research interests lie in (i) Robustness and Safety on Deep Neural Networks; (ii) Machine Learning and Statistical Learning.