Adversarial Robustness Guarantees for Quantum Classifiers
arXiv:2405.10360 · doi:10.1038/s41534-025-01129-3
Abstract
Despite their ever more widespread deployment throughout society, machine learning algorithms remain critically vulnerable to being spoofed by subtle adversarial tampering with their input data. The prospect of near-term quantum computers being capable of running {quantum machine learning} (QML) algorithms has therefore generated intense interest in their adversarial vulnerability. Here we show that quantum properties of QML algorithms can confer fundamental protections against such attacks, in certain scenarios guaranteeing robustness against classically-armed adversaries. We leverage tools from many-body physics to identify the quantum sources of this protection. Our results offer a theoretical underpinning of recent evidence which suggest quantum advantages in the search for adversarial robustness. In particular, we prove that quantum classifiers are: (i) protected against weak perturbations of data drawn from the trained distribution, (ii) protected against local attacks if they are insufficiently scrambling, and (iii) show evidence that they are protected against universal adversarial attacks if they are sufficiently chaotic. Our analytic results are supported by numerical evidence demonstrating the applicability of our theorems and the resulting robustness of a quantum classifier in practice. This line of inquiry constitutes a concrete pathway to advantage in QML, orthogonal to the usually sought improvements in model speed or accuracy.
22 pages, 2 figures. Comments welcome
References in corpus (57)
- Quantum Machine Learning
- Quantum cryptography: Public key distribution and coin tossing
- A bound on chaos
- Barren plateaus in quantum neural network training landscapes
- A Practical Introduction to Tensor Networks: Matrix Product States and Projected Entangled Pair States
- One pixel attack for fooling deep neural networks
- Cost Function Dependent Barren Plateaus in Shallow Parametrized Quantum Circuits
- Fast Scramblers
- Evasion Attacks against Machine Learning at Test Time
- Connecting ansatz expressibility to gradient magnitudes and barren plateaus
- A rigorous and robust quantum speed-up in supervised machine learning
- Measuring the scrambling of quantum information
- Chaos and complexity by design
- Robust data encodings for quantum classifiers
- Lieb-Robinson and the butterfly effect
- Diagnosing Barren Plateaus with Tools from Quantum Optimal Control
- Operator space entanglement entropy in transverse Ising chain
- Barren Plateaus in Variational Quantum Computing
- Does scrambling equal chaos?
- Operator Entanglement in Interacting Integrable Quantum Systems: the Case of the Rule 54 Chain
- Quantum noise protects quantum classifiers against adversaries
- The Eigenstate Thermalization Hypothesis and Out of Time Order Correlators
- A Lie Algebraic Theory of Barren Plateaus for Deep Parameterized Quantum Circuits
- Entanglement Devised Barren Plateau Mitigation
- Quantum Adversarial Machine Learning
- Is efficiency of classical simulations of quantum dynamics related to integrability?
- Entanglement scaling of operators: a conformal field theory approach, with a glimpse of simulability of long-time dynamics in 1+1d
- Barren plateaus preclude learning scramblers
- Introduction to Haar Measure Tools in Quantum Information: A Beginner's Tutorial
- Operator Entanglement in Local Quantum Circuits I: Chaotic Dual-Unitary Circuits
- Experimental quantum adversarial learning with programmable superconducting qubits
- Operator Space Entanglement Entropy in XY Spin Chains
- Accessing scrambling using matrix product operators
- Theoretical Guarantees for Permutation-Equivariant Quantum Neural Networks
- Vulnerability of quantum classification to adversarial perturbations
- Towards quantum enhanced adversarial robustness in machine learning
- Does provable absence of barren plateaus imply classical simulability?
- Scrambling is Necessary but Not Sufficient for Chaos
- Information Scrambling in Quantum Neural Networks
- The Adjoint Is All You Need: Characterizing Barren Plateaus in Quantum Ansätze
- Information scrambling in chaotic systems with dissipation
- Benchmarking Adversarially Robust Quantum Machine Learning at Scale
- Experimental exploration of five-qubit quantum error correcting code with superconducting qubits
- Robust in Practice: Adversarial Attacks on Quantum Machine Learning
- Diffusion and operator entanglement spreading
- Quantum coherence as a signature of chaos
- Robustness Verification of Quantum Classifiers
- Provably Trainable Rotationally Equivariant Quantum Machine Learning
- Trainability Enhancement of Parameterized Quantum Circuits via Reduced-Domain Parameter Initialization
- Scrambling Ability of Quantum Neural Networks Architectures
- Drastic Circuit Depth Reductions with Preserved Adversarial Robustness by Approximate Encoding for Quantum Machine Learning
- Quantum Transfer Learning with Adversarial Robustness for Classification of High-Resolution Image Datasets
- Operational Metric for Quantum Chaos and the Corresponding Spatiotemporal Entanglement Structure
- Training robust and generalizable quantum models
- Quantum neural networks form Gaussian processes
- Defence against adversarial attacks using classical and quantum-enhanced Boltzmann machines
- Bridging Entanglement and Magic Resources within Operator Space