Topics for Interest (included, but not limited to)

• Federated Learning, in particular the challenges security, privacy, robustness, heterogeneous datasets, personalization, fairness, lack of labels (unsupervised learning), resource constraints, aggregation/bandwidth bottleneck, adversarial attacks, scalability, and continual/incremental learning.

• Application domains and benchmarks for trustworthy machine learning, in particular applications in image processing, natural language processing, computer vision, healthcare, IoT, and data sciences.

• New information theoretic and cryptographic approaches for secure and privacy-preserving learning and inference.

• Fairness, Bias, and Trust in decentralized and privacy-preserving machine learning, in particular the challenges of establishing trust in federated learning domains, in which the users and the servers can behave autonomously and can be malicious.

• Hardware security and its role in trustworthy AI

• Multi-Agent AI: In a world with multiple voice services each with different capabilities, the user experience would benefit greatly if these services can “work together” to create a delightful and friction-free experience for the user. The challenge is to improve interoperability between AI’s while preserving user privacy to accomplish a task (for example, if a user initiates a task with agent A and wants to finish the task with agent B, how should the two agents coordinate for task completion), as well as how to take advantage of each assistant’s capabilities in concert to correctly answer a query.

Timeline

Release of CFP: 13 January 2022

Full Proposals Due: 28 April 2022 at 5pm PT

Announcements of Selected Projects: 28 June 2022

New Projects Start Date: 1 July 2022

Access by USC Email Only