Understanding quantum machine learning also requires rethinking generalization

Abstract

Quantum machine learning models have shown successful generalization performance even when trained with few data. In this work, through systematic randomization experiments, we show that traditional approaches to understanding generalization fail to explain the behavior of such quantum models. Our experiments reveal that state-of-the-art quantum neural networks accurately fit random states and random labeling of training data. This ability to memorize random data defies current notions of small generalization error, problematizing approaches that build on complexity measures such as the VC dimension, the Rademacher complexity, and all their uniform relatives. We complement our empirical results with a theoretical construction showing that quantum neural networks can fit arbitrary labels to quantum states, hinting at their memorization ability. Our results do not preclude the possibility of good generalization with few training data but rather rule out any possible guarantees based only on the properties of the model family. These findings expose a fundamental challenge in the conventional understanding of generalization in quantum machine learning and highlight the need for a paradigm shift in the study of quantum models for machine learning tasks.

14+4 pages, 3 figures

References in corpus (7)

• Supplementary information for "Quantum supremacy using a programmable superconducting processor"
• Strong quantum computational advantage using a superconducting quantum processor
• A class of quantum many-body states that can be efficiently simulated
• One-Dimensional Symmetry Protected Topological Phases and their Transitions
• Fantastic Generalization Measures and Where to Find Them
• Decomposition of Matrix Product States into Shallow Quantum Circuits
• Rademacher complexity of noisy quantum circuits

Cited by in corpus (19)

• Does provable absence of barren plateaus imply classical simulability?
• Classically estimating observables of noiseless quantum circuits
• Quantum Convolutional Neural Networks are Effectively Classically Simulable
• A General Approach to Dropout in Quantum Neural Networks
• Training robust and generalizable quantum models
• What can we learn from quantum convolutional neural networks?
• Neural quantum kernels: training quantum kernels with quantum neural networks
• Satellite image classification with neural quantum kernels
• Generalization Error Bound for Quantum Machine Learning in NISQ Era -- A Survey
• Single-shot quantum machine learning
• Data-Dependent Generalization Bounds for Parameterized Quantum Models Under Noise
• Generalization Bounds in Hybrid Quantum-Classical Machine Learning Models
• Quantum Curriculum Learning
• Towards Improved Quantum Machine Learning for Molecular Force Fields
• Double descent in quantum kernel methods
• Transfer entropy and O-information to detect grokking in tensor network multi-class classification problems
• Learning complexity gradually in quantum machine learning models
• Vortex Detection from Quantum Data
• Superior resilience to poisoning and amenability to unlearning in quantum machine learning