Is Feature Selection Secure against Training Data Poisoning?

Learning in adversarial settings is becoming an important task for application domains where attackers may inject malicious data into the training set to subvert normal operation of data-driven techno