Engineering Fair and Efficient Learning-Based Software Systems

The wide adoption of learning-based software systems (LBSS) - i.e., systems that include machine learning models - raises concerns about their quality. Among the most relevant and peculiar quality attributes of these systems is fairness, which identifies a system free from prejudice and discrimination among groups and individuals. At the same time, efficiency, a property shared with traditional software systems, has gained considerable relevance in recent years, especially in the realm of green and sustainable AI. However, the development of fair and efficient LBSS is a challenging and complex task. This talk describes our journey to support the development of fair and efficient LBSS. We present a set of contributions spanning all the phases of a standard development workflow. Those contributions are mainly devoted to democratizing and easing the development of fair and efficient LBSS. Finally, we highlight open challenges that can pave the way for future research in this field.