Angol
Machine learning is widely used in signal processing which is demonstrated in this book. The success of machine learning in signal processing relies heavily on the quality of the data which is also demonstrated in this book. However, the diverse data sources make it harder to get very high-quality data. What makes it worse is that there might be a malicious adversary, who can deliberately modify the data or add poisoning data to corrupt the learning system. This imposes a significant threat to machine learning in signal processing, for example, in wireless communication, array signal processing, and image signal processing. Hence, it is necessary and urgent to investigate the behavior of machine learning algorithms in signal processing under adversarial attacks. The authors examine the adversarial robustness of three commonly used machine learning algorithms in signal processing: linear regression, LASSO-based feature selection, and principal component analysis (PCA). Based on the theoretical analysis for this book, the authors also carry out adversarial attacks on several signal processing problems, such as feature selection, array signal processing, principal component analysis, wireless sensor networks and more. The first part of this book studies the adversarial robustness of linear regression. The authors assume there is an adversary in the linear regression system and it tries to suppress or promote one of the regression coefficients. To obtain this goal, the adversary adds poisoning data samples or directly modifies the feature matrix of the original data. The authors derive the optimal poisoning data sample and propose an alternating optimization method to design the optimal feature modification. It also demonstrate the effectiveness of the attack against a wireless distributed learning system. The second part of this book extends the linear regression to LASSO-based feature selection and studies the best strategy to modify the feature matrix or response values to mislead the learning system to select the wrong features. The authors formulate this problem as a bi-level optimization problem and use a smooth approximation of the norm function to attain the gradient of our objective function. With the gradient information, the authors employ the projected gradient method to find the optimal attacks. It also illustrates how this attack influences array signal processing and weather data analysis. The last section of this book considers the adversarial robustness of the subspace learning problem. The authors examine the optimal modification strategy under the energy constraints to delude the PCA-based subspace learning algorithm and derive the optimal attack strategy to modify the original data to maximize the subspace distance between the original one and the one after modification. This book also conducts an attack on a principal regression problem and demonstrate its impacts on the subspace and the regression result. This book targets researchers working in machine Learning, electronic information and Information theory as well as advanced-level students studying these subjects. R&D engineers who are working in machine learning, adversarial machine learning, robust machine learning and technical consultants working on the security and robustness of machine learning will also want to purchase this book as a reference guide.
