Machine Learning Based Cyber Attack Targeting on Controlled Information

This thesis explore the Machine Learning-based cyber attack targeting on the cotrolled information, especially the training set in Automated Speech Recognition system. Voice interfaces and assistants implemented by various services have become increasingly sophisticated, powered by increased availability of data. The thesis explore the different guarantees about users' audio data enforcing data-protection regulations. In the meanwhile, the acoustic model inherits deep neural networks' vulnerabilities like crafted adversarial examples. This thesis also proposes a novel scheme as an iterative optimization-based method to generate targeted adversarial examples quickly with limited computational resources.