Models of Learning for Quantum Processes - With Noise, Limitations and Adversaries

Explore quantum process learning theory through this 57-minute conference talk that examines realistic models for characterizing quantum systems in the presence of noise and adversarial conditions. Delve into the fundamental problem of learning quantum evolution and discover how sophisticated learning theory techniques can bridge the gap to practical applications in physics and cryptography. Learn about two key physically motivated learning models: statistical query learning of quantum processes, which provides both learning guarantees and robustness to noise, and agnostic process learning, which enables efficient learning even when data sources are noisy or potentially adversarial. Understand how agnostic process learning works by examining scenarios where you have query access to an unknown quantum channel and must output a channel that approximates it as well as the best channel in a known concept class. Discover natural applications of these models in quantum machine learning, quantum metrology, classical simulation, and error mitigation. Gain insights into relevant techniques while exploring open questions and limitations in both quantum process learning models, presented by Mina Doosti at the Quantum Techniques in Machine Learning conference.