Reliable Learning

Reliable learning aims to develop machine learning models that produce provably correct or trustworthy predictions, addressing concerns about model robustness and generalization in challenging environments. Current research focuses on improving model reliability through techniques like conformal prediction, hierarchical reliability propagation, and incorporating constraints directly into model architectures (e.g., using QCQPs). This work is crucial for building trustworthy AI systems across diverse applications, from power grid optimization and cybersecurity to medical diagnosis and robotics, where reliable predictions are paramount for safety and efficacy.