Harnessing AI to Solve Inverse Problems in Quantum Physics

This talk presents novel strategies to tackle a fundamental challenge in quantum mechanics: inferring the underlying potential from electron flow measurements in two-dimensional electron gases (2DEGs). This inverse problem is essential for the design of advanced electronic devices but is complicated by limited datasets, computationally intensive simulations, and a lack of input-output pairs for training artificial intelligence (AI) models.

We propose a machine learning (ML)-based framework to address these challenges and extract the potential landscape of 2DEGs from scanning gate microscopy (SGM) data. By comparing three techniques—image-to-image translation with generative adversarial networks (GANs), cellular neural networks (CNNs), and an evolutionary search algorithm—we demonstrate the effectiveness of ML in identifying and analyzing defects in 2DEGs. Our findings illuminate the interactions between defects and 2DEG properties, paving the way for advancements in quantum computing and nanoelectronics.