Towards Machine Learning Models that We Can Trust: Testing, Improving, and Explaining Robustness

In recent years, machine learning has become the most effective way to analyze massive data streams.However, machine learning is also subject to security and reliability issues.These aspects require machine learning to be thoroughly tested before being deployed in unsupervised scenarios, such as services intended for consumers.The goal of this session is to discuss open challenges, both theoretical and practical, related to the security and safety of machine learning.The session will try to address the following challenges: (i) the implementation of efficient tests for Machine Learning in the context of robustness to attacks and natural drifts of data; and (ii) the design of robust and efficient models able to function in the wild and mitigate or detect adversarial attacks. ContextMachine learning (ML) has rapidly transformed various industries, from powering recommendation systems to driving autonomous vehicles.As ML adoption grows, so does the need for rigorous evaluation and trustworthiness of ML models.In the last years, these models have been rapidly increasing in size and complexity, as well as the amount of data used for their training.This demands for testing techniques able to effectively cover the attack surface of these systems and properly test the resilience of ML to unseen and undesirable attacks [1].To address this open problem, we formulate the following research challenge:Research Challenge 1 Implementation of efficient tests for Machine Learning in the context of robustness to attacks and natural drifts of data.In addition, it is crucial that ML algorithms consider both high technical and functional standards but also additional trustworthy requirements.These models have to be trustworthy in terms of attacks that can come from the misuse of the models and in the meanwhile take into account societal and individual values and principles that significantly impact people's lives, such as ethical concerns.This mandates the design of ML algorithms that we can trust, leading to our second research challenge:Research Challenge 2 Design of robust and efficient models able to function in the wild and mitigate or detect adversarial attacks.