{"id":{"repo_id":"glasgow","oai_identifier":"oai:theses.gla.ac.uk:86191"},"canonical_url":"https://search.ndltd.org/etd/glasgow/oai:theses.gla.ac.uk:86191","repository":{"repo_id":"glasgow","name":"University of Glasgow","base_url":"https://theses.gla.ac.uk/cgi/oai2"},"display":{"title":"From correlation-driven to structure-aware learning: towards innately-intelligent neural networks for wireless systems and beyond","abstract":"Modern wireless networks increasingly rely on artificial intelligence to support spectrum awareness, radio propagation modelling, coverage prediction, and automated network optimisation. Although conventional deep neural networks (DNNs) can achieve strong predictive performance, they are typically correlation-driven, data-hungry, and difficult to interpret. These limitations become critical in wireless systems, where labelled measurements are expensive to obtain, deployment conditions vary across cities, carrier frequencies, antenna configurations, clutter classes, and planning tools, and operational decisions often require traceable physical reasoning rather than opaque predictions. This thesis investigates how wireless learning models can move from purely data-driven prediction toward structure-aware and interpretable intelligence. First, it examines deep learning for wireless spectrum awareness in vehicular communication scenarios, demonstrating the practical value of NN models while also exposing their dependence on learned statistical representations. Building on this limitation, the thesis proposes Innately Intelligent Neural Networks (IINNs), a domain-informed NN modelling framework in which physical variables, analytical relationships, and mathematical operators are embedded directly into the network architecture before training. Unlike post-hoc explainability methods, IINN is designed to be interpretable by construction: intermediate computations correspond to meaningful wireless propagation components. The proposed framework is evaluated for radio propagation and received signal reference power prediction under limited-data and out-of-distribution conditions. Across deployment shifts, IINN demonstrates stronger robustness than representative black-box baselines, including (22.13± 2.91)% relative OOD degradation under joint city–frequency shift compared with 45.32–58.04% for the baseline models. Across training-data fractions from 100% to 5%, IINN achieves the lowest average relative OOD degradation,(17.70±3.38)%. The evaluation also includes propagation label-shift testing, for which absolute OOD RMSE is reported because the target labels are generated by a different propagation model. Structural traceability analysis further shows that IINN exposes learned analytical coefficients and section-wise propagation calculations during inference. Finally, the thesis extends IINN to transfer learning, showing that physically meaningful model components provide a principled basis for deciding which parameters should be retained, adapted, or recalibrated across deployment scenarios. Overall, this thesis positions IINN as a compact, auditable, and data-efficient learning paradigm for wireless systems, particularly suited to settings where partial domain knowledge is available, labelled data are limited, and trustworthy model behaviour is required.","abstract_html":"Modern wireless networks increasingly rely on artificial intelligence to support spectrum awareness, radio propagation modelling, coverage prediction, and automated network optimisation. Although conventional deep neural networks (DNNs) can achieve strong predictive performance, they are typically correlation-driven, data-hungry, and difficult to interpret. These limitations become critical in wireless systems, where labelled measurements are expensive to obtain, deployment conditions vary across cities, carrier frequencies, antenna configurations, clutter classes, and planning tools, and operational decisions often require traceable physical reasoning rather than opaque predictions. This thesis investigates how wireless learning models can move from purely data-driven prediction toward structure-aware and interpretable intelligence. First, it examines deep learning for wireless spectrum awareness in vehicular communication scenarios, demonstrating the practical value of NN models while also exposing their dependence on learned statistical representations. Building on this limitation, the thesis proposes Innately Intelligent Neural Networks (IINNs), a domain-informed NN modelling framework in which physical variables, analytical relationships, and mathematical operators are embedded directly into the network architecture before training. Unlike post-hoc explainability methods, IINN is designed to be interpretable by construction: intermediate computations correspond to meaningful wireless propagation components. The proposed framework is evaluated for radio propagation and received signal reference power prediction under limited-data and out-of-distribution conditions. Across deployment shifts, IINN demonstrates stronger robustness than representative black-box baselines, including (22.13± 2.91)% relative OOD degradation under joint city–frequency shift compared with 45.32–58.04% for the baseline models. Across training-data fractions from 100% to 5%, IINN achieves the lowest average relative OOD degradation,(17.70±3.38)%. The evaluation also includes propagation label-shift testing, for which absolute OOD RMSE is reported because the target labels are generated by a different propagation model. Structural traceability analysis further shows that IINN exposes learned analytical coefficients and section-wise propagation calculations during inference. Finally, the thesis extends IINN to transfer learning, showing that physically meaningful model components provide a principled basis for deciding which parameters should be retained, adapted, or recalibrated across deployment scenarios. Overall, this thesis positions IINN as a compact, auditable, and data-efficient learning paradigm for wireless systems, particularly suited to settings where partial domain knowledge is available, labelled data are limited, and trustworthy model behaviour is required.","abstract_has_math":false,"creators":["Zaidi, Syed Basit Ali"],"institution":"University of Glasgow","degree_name":null,"degree_level":"PhD","degree_discipline":null,"degree_department":null,"school":null,"contributors":[],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2026,"date_issued":"2026","date_published":"2026","updated_at":"2026-11-19T08:37:50Z","subjects":["QA75 Electronic computers. Computer science","TK Electrical engineering. Electronics Nuclear engineering"],"languages":["en"],"rights":[],"rights_urls":[],"identifier_entries":[],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/10.5525/gla.thesis.86191","outbound_label":"DOI","outbound_source":"dc:identifier.doi","record_url":"https://doi.org/10.5525/gla.thesis.86191","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier.doi","download_url":"https://theses.gla.ac.uk/86191/3/2026ZaidiPhD.pdf","download_label":"Download PDF","download_source":"dc:identifier.uri","access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://theses.gla.ac.uk/cgi/oai2?verb=GetRecord&metadataPrefix=uketd_dc&identifier=oai%3Atheses.gla.ac.uk%3A86191","prefix":"uketd_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:creator","label":"Author","values":["Zaidi, Syed Basit Ali"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2026"]},{"key":"dc:date.issued","label":"Date","values":["2026"]},{"key":"dc:publisher.institution","label":"Dc Publisher Institution","values":["University of Glasgow"]},{"key":"dc:relation.isreferencedby","label":"Dc Relation Isreferencedby","values":["https://theses.gla.ac.uk/86191/"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"dc:type.qualificationlevel","label":"Dc Type Qualificationlevel","values":["PhD"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["QA75 Electronic computers. Computer science","TK Electrical engineering. Electronics Nuclear engineering"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["en"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.doi","label":"DOI","values":["10.5525/gla.thesis.86191"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://theses.gla.ac.uk/86191/3/2026ZaidiPhD.pdf"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Modern wireless networks increasingly rely on artificial intelligence to support spectrum awareness, radio propagation modelling, coverage prediction, and automated network optimisation. Although conventional deep neural networks (DNNs) can achieve strong predictive performance, they are typically correlation-driven, data-hungry, and difficult to interpret. These limitations become critical in wireless systems, where labelled measurements are expensive to obtain, deployment conditions vary across cities, carrier frequencies, antenna configurations, clutter classes, and planning tools, and operational decisions often require traceable physical reasoning rather than opaque predictions. This thesis investigates how wireless learning models can move from purely data-driven prediction toward structure-aware and interpretable intelligence. First, it examines deep learning for wireless spectrum awareness in vehicular communication scenarios, demonstrating the practical value of NN models while also exposing their dependence on learned statistical representations. Building on this limitation, the thesis proposes Innately Intelligent Neural Networks (IINNs), a domain-informed NN modelling framework in which physical variables, analytical relationships, and mathematical operators are embedded directly into the network architecture before training. Unlike post-hoc explainability methods, IINN is designed to be interpretable by construction: intermediate computations correspond to meaningful wireless propagation components. The proposed framework is evaluated for radio propagation and received signal reference power prediction under limited-data and out-of-distribution conditions. Across deployment shifts, IINN demonstrates stronger robustness than representative black-box baselines, including (22.13± 2.91)% relative OOD degradation under joint city–frequency shift compared with 45.32–58.04% for the baseline models. Across training-data fractions from 100% to 5%, IINN achieves the lowest average relative OOD degradation,(17.70±3.38)%. The evaluation also includes propagation label-shift testing, for which absolute OOD RMSE is reported because the target labels are generated by a different propagation model. Structural traceability analysis further shows that IINN exposes learned analytical coefficients and section-wise propagation calculations during inference. Finally, the thesis extends IINN to transfer learning, showing that physically meaningful model components provide a principled basis for deciding which parameters should be retained, adapted, or recalibrated across deployment scenarios. Overall, this thesis positions IINN as a compact, auditable, and data-efficient learning paradigm for wireless systems, particularly suited to settings where partial domain knowledge is available, labelled data are limited, and trustworthy model behaviour is required."]},{"key":"dc:format","label":"Dc Format","values":["pdf"]},{"key":"dcterms:accessrights","label":"Dcterms Accessrights","values":["public"]},{"key":"dc:title","label":"Title","values":["From correlation-driven to structure-aware learning: towards innately-intelligent neural networks for wireless systems and beyond"]}]}],"canonical_facts":{"dc:creator":["Zaidi, Syed Basit Ali"],"dc:date":["2026"],"dc:date.issued":["2026"],"dc:description.abstract":["Modern wireless networks increasingly rely on artificial intelligence to support spectrum awareness, radio propagation modelling, coverage prediction, and automated network optimisation. Although conventional deep neural networks (DNNs) can achieve strong predictive performance, they are typically correlation-driven, data-hungry, and difficult to interpret. These limitations become critical in wireless systems, where labelled measurements are expensive to obtain, deployment conditions vary across cities, carrier frequencies, antenna configurations, clutter classes, and planning tools, and operational decisions often require traceable physical reasoning rather than opaque predictions. This thesis investigates how wireless learning models can move from purely data-driven prediction toward structure-aware and interpretable intelligence. First, it examines deep learning for wireless spectrum awareness in vehicular communication scenarios, demonstrating the practical value of NN models while also exposing their dependence on learned statistical representations. Building on this limitation, the thesis proposes Innately Intelligent Neural Networks (IINNs), a domain-informed NN modelling framework in which physical variables, analytical relationships, and mathematical operators are embedded directly into the network architecture before training. Unlike post-hoc explainability methods, IINN is designed to be interpretable by construction: intermediate computations correspond to meaningful wireless propagation components. The proposed framework is evaluated for radio propagation and received signal reference power prediction under limited-data and out-of-distribution conditions. Across deployment shifts, IINN demonstrates stronger robustness than representative black-box baselines, including (22.13± 2.91)% relative OOD degradation under joint city–frequency shift compared with 45.32–58.04% for the baseline models. Across training-data fractions from 100% to 5%, IINN achieves the lowest average relative OOD degradation,(17.70±3.38)%. The evaluation also includes propagation label-shift testing, for which absolute OOD RMSE is reported because the target labels are generated by a different propagation model. Structural traceability analysis further shows that IINN exposes learned analytical coefficients and section-wise propagation calculations during inference. Finally, the thesis extends IINN to transfer learning, showing that physically meaningful model components provide a principled basis for deciding which parameters should be retained, adapted, or recalibrated across deployment scenarios. Overall, this thesis positions IINN as a compact, auditable, and data-efficient learning paradigm for wireless systems, particularly suited to settings where partial domain knowledge is available, labelled data are limited, and trustworthy model behaviour is required."],"dc:format":["pdf"],"dc:identifier.doi":["10.5525/gla.thesis.86191"],"dc:identifier.uri":["https://theses.gla.ac.uk/86191/3/2026ZaidiPhD.pdf"],"dc:language":["en"],"dc:publisher.institution":["University of Glasgow"],"dc:relation.isreferencedby":["https://theses.gla.ac.uk/86191/"],"dc:subject":["QA75 Electronic computers. Computer science","TK Electrical engineering. Electronics Nuclear engineering"],"dcterms:accessrights":["public"],"dc:title":["From correlation-driven to structure-aware learning: towards innately-intelligent neural networks for wireless systems and beyond"],"dc:type":["Thesis"],"dc:type.qualificationlevel":["PhD"]},"updated_at":"2026-11-19T08:37:50Z"}