‘Protecting AI Solutions From Attacks’ at ML Conference Online

ML Conference brings together the world’s leading Machine Learning experts and innovators to share their ideas and experience. The founder and CEO of Adversa Alex Polyakov became a part of this event with his presentation called ‘Protecting AI Solutions From Attacks’. Attacks on machine learning systems include a wide range of different approaches and don’t end with the notorious Adversarial examples. Attacks can change the logic of the system (adversarial examples and reprogramming) to obtain data from AI systems (so-called Membership inference or Model Extraction attacks) or, conversely, to inject data into the system (Poisoning, Backdoor, Trojan). The silver bullet from these attacks hasn’t been invented and is unlikely to be.  Alexander demonstrated 

• how to approach the security assessment of AI algorithms correctly; 
• what metrics to look at; 
• which protection can be applied;
• where is the best place to apply; 
• how to eventually get the maximum protection for the reasonable investment of resources.