Sparsified Kolmogorov-Arnold Networks for Interpretable Quantum State Tomography

2026-06-10 · Source: Artificial Intelligence · Field: Science & Research — Physical Sciences & Chemistry, Artificial Intelligence & Machine Learning, Mathematics & Computational Sciences · Depth: Expert, quick

Summary

Sparsified Kolmogorov-Arnold Networks (KANs) are explored for quantum state tomography, aiming for both high reconstruction fidelity and an inspectable physical structure. Researchers applied a KAN to a controlled three-qubit GHZ-family benchmark, using all 63 non-identity Pauli expectation values to reconstruct three GHZ-subspace variables: population imbalance $z$, real off-diagonal component $c$, and imaginary off-diagonal component $s$. Under finite-shot sampling and depolarizing noise, the KAN successfully identified the extended 12-channel GHZ-relevant Pauli set from the 63 measurements, achieving exact top-12 recovery across various shot counts and noise strengths. These support patterns remained stable across multi-seed random initializations and noise levels, collapsing only under random-label controls. The dominant pruned input-hidden-output pathways organized Z-type population observables and X/Y off-diagonal observables in a manner consistent with the analytic GHZ Pauli grouping, and sparse formula recovery matched canonical signed Pauli relations. The KAN's primary contribution is pathway-level structural interpretability within a neural reconstruction model.

Key takeaway

For research scientists developing machine learning models for quantum state tomography, you should consider sparsified Kolmogorov-Arnold Networks (KANs) to ensure interpretability. KANs offer pathway-level structural insights, allowing you to audit learned reconstruction rules against known physical structures like Pauli groupings. This approach provides a consistency chain, verifying that your model's internal organization aligns with established quantum mechanics, which is crucial for building trust and understanding in complex quantum ML applications.

Key insights

Sparsified KANs offer structural interpretability for quantum state tomography by revealing underlying physical structures.

Principles

• KANs can reveal physical structure in learned models.
• Ablation identifies relevant features in noisy data.
• Consistency checks audit learned rules.

Method

A sparsified KAN is trained on Pauli expectation values to reconstruct quantum state variables, then ablated and analyzed for pathway organization and sparse formula recovery.

In practice

• Audit learned quantum reconstruction rules against known physics.
• Use KANs to identify critical measurement sets.
• Apply KANs for interpretable ML in quantum systems.

Topics

• Quantum State Tomography
• Kolmogorov-Arnold Networks
• Machine Learning Interpretability
• Pauli Observables
• Quantum Information
• Sparsification

Best for: AI Scientist, Research Scientist

Related on AIssential

• A Practitioner's Guide to Kolmogorov-Arnold Networks — KANs offer enhanced expressivity and interpretability over MLPs by using learnable univariate basis functions on network edges.
• Provable and scalable quantum Gaussian processes for quantum learning — Quantum Gaussian processes offer a scalable, interpretable Bayesian framework for learning from quantum data.
• The Hidden Bottleneck in Quantum Machine Learning: Getting Data into a Quantum Computer — Efficiently loading classical data into quantum systems is a critical, unsolved bottleneck in Quantum Machine Learning.
• Equilibrium Propagation and Hamiltonian Inference in the Diffusive Fitzhugh-Nagumo Model — Equilibrium Propagation and Hamiltonian Echo Backpropagation can be extended to skew-gradient systems and deep Energy-Based Models for efficient credit assignment and inference.
• Conceptualizing Embeddings: Sparse Disentanglement for Vision-Language Models — CEDAR disentangles vision-language model embeddings post-hoc, revealing compositional structure without increasing dimensionality, improving interpretability.
• I-SAFE: Wasserstein Coherence Metrics for Structural Auditing of Scientific AI Models — Scientific AI model performance benchmarks are insufficient to confirm alignment with domain-relevant structural behavior.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence.