Metraipolis › archive › dc_226c8014feff

Watching Quantum Models Think: Hilbert-Space Interpretability in Quantum Transformer Blocks — arXiv:2609.23016

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.23016
License (as archived): CC0-1.0
Archived: 2026-10-09T07:08:20.875Z
SHA-256 of the archived text: 2489a398a898dd907b531ca1cc2ccbe9dfbbd9e89f71507627ac22767aa19da8
Stable URL of this page: https://metraipolis.com/documento/dc_226c8014feff

How to cite

Watching Quantum Models Think: Hilbert-Space Interpretability in Quantum Transformer Blocks — arXiv:2609.23016 — arXiv (en), https://arxiv.org/abs/2609.23016 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_226c8014feff (Metraipolis archive dc_226c8014feff, sha256 2489a398a898dd907b531ca1cc2ccbe9dfbbd9e89f71507627ac22767aa19da8)

BibTeX

@misc{dc226c8014feff,
  title = {Watching Quantum Models Think: Hilbert-Space Interpretability in Quantum Transformer Blocks — arXiv:2609.23016},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.23016},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_226c8014feff; sha256 2489a398a898dd907b531ca1cc2ccbe9dfbbd9e89f71507627ac22767aa19da8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.23016. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Deep learning models are powerful but opaque. As quantum machine learning matures, the field faces a defining choice: build quantum models that are equally opaque, or exploit the mathematical structure of quantum mechanics to make them inherently interpretable. We show that the latter is possible. By tracking quantum mutual information~(MI), entanglement entropy, and state fidelity through the layers of a Quantum Transformer Block (\qtb{}), a fully-coherent variational circuit with quantum analogues of both attention and feedforward, we gain direct insight into how the model processes information: which tokens it attends to, when correlations form, and why predictions fail. On four tasks with known dependency structure we show that (i)~learned MI matrices align with ground-truth task structure (AUC$\,{=}\,0.69$ on lookup), (ii)~disabling entangling gates collapses accuracy from 100\% to 15\% while MI$\to 0$, proving entanglement is the mechanism, (iii)~accuracy and MI co-evolve during training ($\rho\,{=}\,0.92$ on lookup), and (iv)~per-sample MI predicts prediction correctness on the conditional task with ROC AUC$\,{=}\,0.84$. All results are validated on IBM Quantum hardware (ibm\_kingston, Heron~r2): the circuit's reasoning process, from product state through structured entanglement, is directly observable on a superconducting processor. These proof-of-concept results, obtained on small synthetic tasks, suggest that the physics of quantum computation can provide intrinsic interpretability signals with no direct classical counterpart, motivating study of whether this advantage persists at scale.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.LG).
Categorie arXiv: Quantum Physics (quant-ph) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.23016 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-09T06:58:09.544Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).