By Anindita Maiti, Perimeter Institute for Theoretical Physics

Abstract 

 Despite rapid progress in the performance of State-of-the-Art AI models in quantum, most such methods remain black boxes: lacking guarantees of robust and reliable predictions that meet uncertainty quantification benchmarks essential in scientific domains. To address this gap, I will present a few directions that improve robustness, mechanistic interpretability, and uncertainty quantification of complex learning and sample generation abilities. First, I will present the simplest model capable of in-context learning, an ability that underpins LLM success, especially for quantum, for the simplest class of tasks: linear regression. The model performance is exactly derived in the joint asymptotic limit of a large number of samples, token dimensions, sample length, and task diversity: exhibiting a phase transition from memorization to generalization, and is supported by experiments on Transformers. Time permitting, I will next present a Quantum Informed Neural Network paradigm as a quantum sample generation strategy alternative to quantum computers. A tool for systematic coarse-graining of data features irrelevant to learning, based on Renormalization Group (RG), a canon of quantum physics, is introduced, which improves uncertainty quantification. Altogether, these works advance AI reliability in a first-principles manner, while bridging AI with fundamental physics.