Adversarial Attacks in Machine Learning

Machine Learning, especially its branch Deep Learning, is rapidly transforming myriad aspects of our lives: our daily routines, our financial systems - even our healthcare. Deep neural networks, the basis of deep learning,have shown success beyond human capabilities in solving complex problems for different applications ranging from medical diagnoses to self-driving cars, from face recognition systems to financial stock predictions.However, it was recently it was shown that adversaries can easily behave in malicious way against machine learning systems, compromising people’s confidence in them, questioning both security and privacy aspects of such applications. The topic is very interesting and reasonably needs more scientific studies on it. In this work,the focus will be given for three topics. In this thesis, three different pojects on adversarial attacks in machine learning will be considered.