Quantum
Certifying quantum generative models beyond the Fisher trace
Publication: quant-ph updates on arXiv.org
Author: Vicente Leyton-Ortega, Kathleen Hamilton
Source published: 17/11/2026, 05:00:00 UTC
Briefing created: 17/11/2026, 08:41:53 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 17/11/2026, 08:28:25 UTC.
Briefings
A new arXiv preprint (2611.18955v1) analyzes quantum generative models using Fisher-trace and kernel-based certification. It finds that for real-amplitude circuits built from half-angle rotations, classical Fisher information equals quantum Fisher information, making the trace uninformative about fit quality. The Fisher eigenvalue rank counts live nodes in the amplitude tree, which is uninformative on full-support benchmarks. The pooled Wasserstein and maximum-mean-discrepancy scores shift with optimizer convergence and scoring conventions. For two bitstring defect families, the discrepancy vote rejects generators only at certification-grade tolerances, opening only as qubit numbers grow. On Hamming cost, a sharp tolerance ratio exists above which the vote is redundant, with an exponentially small worst-parity minimax value. A numerical campaign at these tolerances showed the discrepancy vote changed no verdict at a thousand shots or more. The false-pass guarantee does not extend to device noise.
Key points
- For real-amplitude quantum circuits built from half-angle rotations, classical Fisher information equals quantum Fisher information almost everywhere, meaning the trace equals the parameter count and carries no information about fit quality.
- The Fisher eigenvalue rank counts the live nodes of the amplitude tree, which coincides with the label on a sparse benchmark but is uninformative on a full-support one.
Context
This research addresses certification for quantum generative models beyond the Fisher trace, analyzing real-amplitude quantum circuit Born machines in noiseless statevector simulation. The findings clarify when classical and quantum metrics align and when discrepancy voting fails, with implications for verifying generated quantum states.
Limitations
Evidence is limited to noiseless statevector simulation; results do not extend to device noise. Numerical campaigns at specified tolerances showed no verdict change at a thousand shots or more. Closed-form solutions exist only for two specific bitstring defect families, limiting generalizability.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 17/11/2026, 08:41:53 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeif63wpl35oz4dcvhyhe3wbywof7ixoy2gmved6zppbd6qcqhztno4