A technique borrowed from fields like image synthesis and protein structure prediction is now being applied to the challenges of quantum computing, as Lino Hofstetter, a visitor to the Computer Science and Technology Department at Cambridge from ETH Zurich, presents a new tutorial on denoising diffusion probabilistic models. These models function by intentionally adding noise to data and then training a neural network to reverse that process, a method Hofstetter will explain as a foundation for generating and optimizing quantum circuits. This talk marks a change for the “Quantum Computer Architectures and Theory” seminar series, which is adding tutorials to its established format. Hofstetter’s presentation explains how these models learn to sample from complex data distributions by reversing a fixed noise corruption process, potentially unlocking new approaches to quantum circuit synthesis and related tasks.
Denoising Diffusion Models for Generative Data Sampling
Hofstetter’s tutorial, delivered at the Computer Laboratory’s William Gates Building, explains how these models can be utilized to generate, optimize, and even discover quantum circuits, laying the conceptual groundwork for their use in quantum circuit synthesis. A core concept involves a Markov chain that progressively transforms data into Gaussian noise, with a neural network then trained to approximate the reverse transitions by predicting the noise added at each step. At sampling time, the model initiates from pure noise and iteratively applies these learned denoising steps to generate structured samples. This tutorial, accessible via Teams, offers a perspective on leveraging generative models for quantum tasks, potentially streamlining the development of complex quantum algorithms and circuits. The presentation details how these models can move beyond traditional methods, offering a new pathway for innovation in quantum computation.
Hofstetter’s tutorial, accessible via Teams, focuses on the conceptual foundations of using these models for quantum circuit synthesis and related tasks, offering a new avenue for exploration within the field.
While diffusion models have achieved strong results in areas such as image synthesis and protein structure prediction, this tutorial focuses on their application to quantum computing.
