ExplainabilityTransparencyAccountabilityquant-ph
eXplainable AI for Quantum Machine Learning
Patrick Steinmüller, Tobias Schulz, Ferdinand Graf, Daniel Herr
November 2, 2022
arXiv.org
20 citations
3 influential
Key Takeaways
-
Quantum machine learning systems pose unique challenges for explainability: Unlike traditional AI, Parametrized Quantum Circuits introduce probabilistic errors from quantum measurements that can destabilize existing explanation methods, making it harder to understand why these systems make specific decisions.
-
Computational complexity grows exponentially with scale: The paper identifies that quantum systems' phase space expands exponentially with each additional qubit, meaning standard explainability techniques like SHAP and Integrated Gradients become computationally impractical as quantum ML systems grow larger—a critical barrier to real-world deployment.
-
Established explainability methods need adaptation for quantum systems: The research evaluates how well-known xAI approaches (Baseline SHAP and Integrated Gradients) perform on quantum circuits, revealing that these methods designed for classical AI don't directly transfer to quantum computing without modification.
-
Quantum mechanics can be leveraged to improve efficiency: Rather than treating quantum properties as purely obstacles, the paper explores how the internal mechanics of quantum circuits themselves can be used to speed up explainability computations, suggesting a path forward for making quantum ML systems more interpretable.
Abstract
Parametrized Quantum Circuits (PQCs) enable a novel method for machine learning (ML). However, from a computational point of view they present a challenge to existing eXplainable AI (xAI) methods. On the one hand, measurements on quantum circuits introduce probabilistic errors which impact the convergence of these methods. On the other hand, the phase space of a quantum circuit expands exponentially with the number of qubits, complicating efforts to execute xAI methods in polynomial time. In this paper we will discuss the performance of established xAI methods, such as Baseline SHAP and Integrated Gradients. Using the internal mechanics of PQCs we study ways to speed up their computation.