This workshop, sponsored by AIM and the NSF, will advance mathematically rigorous methods for fairness and privacy in machine learning and deepen the mathematical understanding of the underlying problems. One thrust of the workshop will advance algorithmic methods to detect and mitigate bias, including deeper study of how embeddings represent topics and potentially propagate bias. Motivated by privacy regulations and the need to remove data influence without retraining, a second thrust focuses on machine unlearning, covering efficient algorithms and provable certification, with strategies for underspecified data. A third thrust will focus on differential privacy in fair ML.

The workshop aims to seed new collaborations and foster a community of researchers at the interface of mathematics, ML foundations, fairness, privacy, and unlearning. The main topics for the workshop are:

• Algorithmic and mathematical foundations of fairness in ML.
• Algorithmic and mathematical foundations of machine unlearning.
• Differential privacy and its tradeoffs with other desired properties of ML systems, such as fairness.