Deep Quantum Graph Dreaming: Deciphering Neural Network Insights into Quantum Experiments

Abstract

Despite their promise to facilitate new scientific discoveries, the opaqueness of neural networks presents a challenge in interpreting the logic behind their findings. Here, we use a eXplainable-AI (XAI) technique called $inception$ or $deep$ $dreaming$, which has been invented in machine learning for computer vision. We use this technique to explore what neural networks learn about quantum optics experiments. Our story begins by training deep neural networks on the properties of quantum systems. Once trained, we "invert" the neural network -- effectively asking how it imagines a quantum system with a specific property, and how it would continuously modify the quantum system to change a property. We find that the network can shift the initial distribution of properties of the quantum system, and we can conceptualize the learned strategies of the neural network. Interestingly, we find that, in the first layers, the neural network identifies simple properties, while in the deeper ones, it can identify complex quantum structures and even quantum entanglement. This is in reminiscence of long-understood properties known in computer vision, which we now identify in a complex natural science task. Our approach could be useful in a more interpretable way to develop new advanced AI-based scientific discovery techniques in quantum physics.

arXiv commentModified Figure 2. Fixed minor typos

References in corpus (33)

• Adam: A Method for Stochastic Optimization
• Machine learning and the physical sciences
• A Survey on Explainable Artificial Intelligence (XAI): Towards Medical XAI
• Explainable Artificial Intelligence: Understanding, Visualizing and Interpreting Deep Learning Models
• A Survey on the Explainability of Supervised Machine Learning
• Discovering physical concepts with neural networks
• On scientific understanding with artificial intelligence
• Automated Search for new Quantum Experiments
• Bell's theorem with and without inequalities for the three-qubit Greenberger-Horne-Zeilinger and W states
• Artificial Intelligence and Machine Learning for Quantum Technologies
• Entanglement by Path Identity
• Machine Learning of Explicit Order Parameters: From the Ising Model to SU(2) Lattice Gauge Theory
• Probing many-body localization with neural networks
• Machine learning method for state preparation and gate synthesis on photonic quantum computers
• Equivariant Diffusion for Molecule Generation in 3D
• Quantum Experiments and Graphs: Multiparty States as coherent superpositions of Perfect Matchings
• Computer-inspired Quantum Experiments
• Computer-inspired concept for high-dimensional multipartite quantum gates
• Unsupervised machine learning of topological phase transitions from experimental data
• Learning hard quantum distributions with variational autoencoders
• Discovering Symmetry Invariants and Conserved Quantities by Interpreting Siamese Neural Networks
• Quantum Experiments and Graphs II: Quantum Interference, Computation and State Generation
• Multiphoton non-local quantum interference controlled by an undetected photon
• Quantum Experiments and Graphs III: High-Dimensional and Multi-Particle Entanglement
• Conceptual understanding through efficient inverse-design of quantum optical experiments
• On-chip quantum interference between the origins of a multi-photon state
• Phase Detection with Neural Networks: Interpreting the Black Box
• Digital Discovery of 100 diverse Quantum Experiments with PyTheus
• Learning Interpretable Representations of Entanglement in Quantum Optics Experiments using Deep Generative Models
• Automated design of quantum optical experiments for device-independent quantum key distribution
• Explainable Representation Learning of Small Quantum States
• Experimental Solutions to the High-Dimensional Mean King's Problem
• Quantum Deep Dreaming: A Novel Approach for Quantum Circuit Design