Quantum Annealing and its Applications to AI Training

Quantum annealers take a different path from gate-based quantum computers: rather than running circuits of gates, they let a network of interacting spins relax toward low-energy states. In this talk, I discuss principles and applications of quantum annealing. The main focus will be AI training. Using a simple image classification task, I will show how the annealer can serve as a neural network and discuss whether quantum effects and tailored annealing protocols could speed up its training. I also discuss some open problems from our work, including grokking: the sudden shift from memorizing training data to genuinely learning from it. Grokking has been observed widely in classical neural network architectures, and now we are trying to reproduce it with our spin system.