{"id":27018,"date":"2026-09-27T20:02:28","date_gmt":"2026-09-27T18:02:28","guid":{"rendered":"https:\/\/www.itspy.cz\/thesis\/machine-learning-models-for-quantum-state-tomography\/"},"modified":"2026-09-27T20:05:07","modified_gmt":"2026-09-27T18:05:07","slug":"machine-learning-models-for-quantum-state-tomography","status":"publish","type":"thesis","link":"https:\/\/www.itspy.cz\/en\/thesis\/machine-learning-models-for-quantum-state-tomography\/","title":{"rendered":"Machine learning models for quantum state tomography"},"content":{"rendered":"<p>This thesis asks whether modern machine-learning model<br \/>\nfamilies and restoration ideas that have driven progress in<br \/>\ncomputer graphics can also improve quantum state<br \/>\ntomography, the reconstruction of an unknown quantum state<br \/>\nfrom repeated measurements. We study four broad families \u2014<br \/>\nfeed-forward neural networks, conditional generative<br \/>\nadversarial networks, iterative generative models from<br \/>\ndiffusion to flow matching, and transformers \u2014 across three<br \/>\npractical regimes: incomplete near-clean measurements,<br \/>\ncomplete finite-shot measurements, and measurements that<br \/>\nare both incomplete and noisy. Five selected<br \/>\npoint-estimation pipelines are developed through<br \/>\nvalidation-guided experiments and evaluated on the same<br \/>\nsynthetic two-qubit state partitions and paired test<br \/>\nrecords. The common formulation treats missing measurements<br \/>\nas masked data, represents shot-dependent reliability<br \/>\nthrough confidence, and uses a normalised Cholesky map to<br \/>\nensure a physical output. Combined with set-based<br \/>\naggregation and Bures-based training, this formulation<br \/>\nreduces the complete high-shot mean Bures distance of a<br \/>\nnaive multilayer perceptron from 0.079 to 0.022. On this<br \/>\ncommon in-distribution benchmark, RhoTransformer has the<br \/>\nlowest mean Bures distance at all four headline operating<br \/>\npoints, reaching 0.015 with all sixteen measurements at<br \/>\n10^5 shots per measurements and 0.103 with twelve<br \/>\nmeasurements at 10^4 shots per observed measurement. A<br \/>\ncontrolled decoder ablation further finds that query self<br \/>\nattention improves complete high-shot reconstruction<br \/>\nrelative to a larger attention-free con-<br \/>\ntrol. Together, the selected-pipeline ranking and decoder<br \/>\nablation raise an intriguing possibility: the \u201cattention is<br \/>\nall you need\u201d idea may extend to quantum<br \/>\nstate tomography as well.<\/p>\n","protected":false},"featured_media":26799,"template":"","meta":{"_acf_changed":false,"_links_to":"","_links_to_target":""},"university":[465,426],"thesis-year":[466],"class_list":["post-27018","thesis","type-thesis","status-publish","has-post-thumbnail","hentry","thesis-year-466"],"acf":{"autor":"Marko \u010cechovi\u010d","portret":"","vedouci":"RNDr. Pavel Bal\u00e1\u017e, Ph.D."},"_links":{"self":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/27018","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis"}],"about":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/types\/thesis"}],"version-history":[{"count":1,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/27018\/revisions"}],"predecessor-version":[{"id":27073,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis\/27018\/revisions\/27073"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/media\/26799"}],"wp:attachment":[{"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/media?parent=27018"}],"wp:term":[{"taxonomy":"university","embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/university?post=27018"},{"taxonomy":"thesis-year","embeddable":true,"href":"https:\/\/www.itspy.cz\/en\/wp-json\/wp\/v2\/thesis-year?post=27018"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}