{"id":{"repo_id":"vt","oai_identifier":"oai:vtechworks.lib.vt.edu:10919/140660"},"canonical_url":"https://search.dev.ndltd.org/etd/vt/oai:vtechworks.lib.vt.edu:10919/140660","repository":{"repo_id":"vt","name":"Virginia Tech","base_url":"https://vtechworks.lib.vt.edu/oai/request"},"display":{"title":"Interpretability and Debugging for Distributed Privacy Preserving Machine Learning","abstract":"Machine learning systems increasingly rely on privacy-preserving distributed training to leverage sensitive data across multiple organizations without centralization. Federated Learning (FL), a distributed privacy-preserving machine learning paradigm, enables hospitals, devices, and enterprises to collaboratively train models without accessing raw client data (e.g., Siri, Alexa, and healthcare applications). Centralized machine learning benefits from rich debugging and interpretability techniques enabled by transparent access to training data. However, FL removes this transparency, rendering traditional techniques ineffective and making debugging and interpretability a challenging open problem. This thesis addresses this challenge by asking: How can we design automated debugging and interpretability methods for federated learning that effectively localize faults and attribute global model predictions without degrading performance or violating FL's core privacy principles? The central insight is that effective debugging and interpretability can be achieved by analyzing model parameters, activations, and gradients-information already shared or derivable in standard FL protocols (e.g., FedAvg). We present three contributions. First, towards fault localization, we redesign traditional differential testing to operate on neuron activations produced by auto-generated inputs, exploiting the fact that faulty clients produce models with divergent activations. Second, we introduce neuron provenance, which decouples data-influence tracking from data access. It identifies influential neurons via gradient-based weighting and decomposes them to client-specific origins, yielding ranked lists of responsible clients across CNNs and Transformers. Third, we extend neuron provenance to federated LLMs, where autoregressive generation and billion-parameter scale make naive tracking infeasible. It introduces token-level provenance at targeted transformer layers, achieving high attribution accuracy across multiple LLM architectures. In each case, the solution operates entirely on information available at the aggregator, requiring no client-side instrumentation. Collectively, these contributions culminate in practical tools that integrate seamlessly with existing distributed ML workflows, enabling real-time debugging and transparent model insights for both classification and LLMs in FL.","abstract_html":"Machine learning systems increasingly rely on privacy-preserving distributed training to leverage sensitive data across multiple organizations without centralization. Federated Learning (FL), a distributed privacy-preserving machine learning paradigm, enables hospitals, devices, and enterprises to collaboratively train models without accessing raw client data (e.g., Siri, Alexa, and healthcare applications). Centralized machine learning benefits from rich debugging and interpretability techniques enabled by transparent access to training data. However, FL removes this transparency, rendering traditional techniques ineffective and making debugging and interpretability a challenging open problem. This thesis addresses this challenge by asking: How can we design automated debugging and interpretability methods for federated learning that effectively localize faults and attribute global model predictions without degrading performance or violating FL's core privacy principles? The central insight is that effective debugging and interpretability can be achieved by analyzing model parameters, activations, and gradients-information already shared or derivable in standard FL protocols (e.g., FedAvg). We present three contributions. First, towards fault localization, we redesign traditional differential testing to operate on neuron activations produced by auto-generated inputs, exploiting the fact that faulty clients produce models with divergent activations. Second, we introduce neuron provenance, which decouples data-influence tracking from data access. It identifies influential neurons via gradient-based weighting and decomposes them to client-specific origins, yielding ranked lists of responsible clients across CNNs and Transformers. Third, we extend neuron provenance to federated LLMs, where autoregressive generation and billion-parameter scale make naive tracking infeasible. It introduces token-level provenance at targeted transformer layers, achieving high attribution accuracy across multiple LLM architectures. In each case, the solution operates entirely on information available at the aggregator, requiring no client-side instrumentation. Collectively, these contributions culminate in practical tools that integrate seamlessly with existing distributed ML workflows, enabling real-time debugging and transparent model insights for both classification and LLMs in FL.","abstract_has_math":false,"creators":["Gill, Waris"],"institution":"Virginia Tech","degree_name":"Doctor of Philosophy","degree_level":"doctoral","degree_discipline":"Computer Science & Applications","degree_department":"Computer Science and#38; Applications","school":null,"contributors":[],"advisors":[],"committee_chairs":["Gulzar, Muhammad Ali"],"committee_members":["Viswanath, Bimal","Ji, Bo","Tilevich, Eli","Anwar, Ali"],"year":2026,"date_issued":"2026-01-07","date_published":"2026-01-07","updated_at":"2026-07-22T22:20:03Z","subjects":["Federated Learning","Privacy","Interpretability","Debugging","LLMs","CNNs"],"languages":["en"],"rights":["In Copyright"],"rights_urls":["http://rightsstatements.org/vocab/InC/1.0/"],"identifier_entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:45318"],"render_values":[{"text":"vt_gsexam:45318","href":null,"code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/10919/140660","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.committeechair","label":"Committee Chair","values":["Gulzar, Muhammad Ali"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Viswanath, Bimal","Ji, Bo","Tilevich, Eli","Anwar, Ali"]},{"key":"dc:contributor.department","label":"Department","values":["Computer Science and#38; Applications"]},{"key":"dc:creator","label":"Author","values":["Gill, Waris"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-01-08T09:00:30Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2026-01-08T09:00:30Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-01-07"]},{"key":"dc:publisher","label":"Institution","values":["Virginia Tech"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Computer Science & Applications"]},{"key":"thesis:degree_level","label":"Degree Level","values":["doctoral"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["Virginia Polytechnic Institute and State University"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Federated Learning","Privacy","Interpretability","Debugging","LLMs","CNNs"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["In Copyright"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://rightsstatements.org/vocab/InC/1.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:45318"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10919/140660"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Machine learning systems increasingly rely on privacy-preserving distributed training to leverage sensitive data across multiple organizations without centralization. Federated Learning (FL), a distributed privacy-preserving machine learning paradigm, enables hospitals, devices, and enterprises to collaboratively train models without accessing raw client data (e.g., Siri, Alexa, and healthcare applications). Centralized machine learning benefits from rich debugging and interpretability techniques enabled by transparent access to training data. However, FL removes this transparency, rendering traditional techniques ineffective and making debugging and interpretability a challenging open problem. This thesis addresses this challenge by asking: How can we design automated debugging and interpretability methods for federated learning that effectively localize faults and attribute global model predictions without degrading performance or violating FL's core privacy principles? The central insight is that effective debugging and interpretability can be achieved by analyzing model parameters, activations, and gradients-information already shared or derivable in standard FL protocols (e.g., FedAvg). We present three contributions. First, towards fault localization, we redesign traditional differential testing to operate on neuron activations produced by auto-generated inputs, exploiting the fact that faulty clients produce models with divergent activations. Second, we introduce neuron provenance, which decouples data-influence tracking from data access. It identifies influential neurons via gradient-based weighting and decomposes them to client-specific origins, yielding ranked lists of responsible clients across CNNs and Transformers. Third, we extend neuron provenance to federated LLMs, where autoregressive generation and billion-parameter scale make naive tracking infeasible. It introduces token-level provenance at targeted transformer layers, achieving high attribution accuracy across multiple LLM architectures. In each case, the solution operates entirely on information available at the aggregator, requiring no client-side instrumentation. Collectively, these contributions culminate in practical tools that integrate seamlessly with existing distributed ML workflows, enabling real-time debugging and transparent model insights for both classification and LLMs in FL."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["When multiple organizations like hospitals, smartphone companies, or banks want to train machine learning models together, they face a dilemma: combining their data would produce better AI, but sharing sensitive patient records, personal messages, or financial information is often illegal or dangerous. Federated Learning, a privacy-preserving approach to machine learning, solves this by allowing organizations to train AI collaboratively without ever sharing their private data. This technology already powers voice assistants like Siri and Alexa. However, this privacy comes with a hidden cost. When something goes wrong, such as when an AI makes a mistake or produces harmful outputs, developers cannot investigate because they cannot see the original data. Traditional debugging and interpretability approaches, designed for centralized data access, are not directly applicable in this distributed, privacy-constrained setting. This thesis develops new methods that allow developers to identify which organization caused a problem and understand why the AI behaves as it does, all without ever accessing private data. We introduce techniques that work across different types of AI, from image recognition systems to advanced language models. These contributions, now integrated into widely-used federated learning platforms, make privacy-preserving AI more trustworthy and deployable in sensitive domains like healthcare and finance."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["Doctor of Philosophy"]},{"key":"dc:format.medium","label":"Dc Format Medium","values":["ETD"]},{"key":"dc:title","label":"Title","values":["Interpretability and Debugging for Distributed Privacy Preserving Machine Learning"]}]}],"canonical_facts":{"dc:contributor.committeechair":["Gulzar, Muhammad Ali"],"dc:contributor.committeemember":["Viswanath, Bimal","Ji, Bo","Tilevich, Eli","Anwar, Ali"],"dc:contributor.department":["Computer Science and#38; Applications"],"dc:creator":["Gill, Waris"],"dc:date.accessioned":["2026-01-08T09:00:30Z"],"dc:date.available":["2026-01-08T09:00:30Z"],"dc:date.issued":["2026-01-07"],"dc:description.abstract":["Machine learning systems increasingly rely on privacy-preserving distributed training to leverage sensitive data across multiple organizations without centralization. Federated Learning (FL), a distributed privacy-preserving machine learning paradigm, enables hospitals, devices, and enterprises to collaboratively train models without accessing raw client data (e.g., Siri, Alexa, and healthcare applications). Centralized machine learning benefits from rich debugging and interpretability techniques enabled by transparent access to training data. However, FL removes this transparency, rendering traditional techniques ineffective and making debugging and interpretability a challenging open problem. This thesis addresses this challenge by asking: How can we design automated debugging and interpretability methods for federated learning that effectively localize faults and attribute global model predictions without degrading performance or violating FL's core privacy principles? The central insight is that effective debugging and interpretability can be achieved by analyzing model parameters, activations, and gradients-information already shared or derivable in standard FL protocols (e.g., FedAvg). We present three contributions. First, towards fault localization, we redesign traditional differential testing to operate on neuron activations produced by auto-generated inputs, exploiting the fact that faulty clients produce models with divergent activations. Second, we introduce neuron provenance, which decouples data-influence tracking from data access. It identifies influential neurons via gradient-based weighting and decomposes them to client-specific origins, yielding ranked lists of responsible clients across CNNs and Transformers. Third, we extend neuron provenance to federated LLMs, where autoregressive generation and billion-parameter scale make naive tracking infeasible. It introduces token-level provenance at targeted transformer layers, achieving high attribution accuracy across multiple LLM architectures. In each case, the solution operates entirely on information available at the aggregator, requiring no client-side instrumentation. Collectively, these contributions culminate in practical tools that integrate seamlessly with existing distributed ML workflows, enabling real-time debugging and transparent model insights for both classification and LLMs in FL."],"dc:description.abstractgeneral":["When multiple organizations like hospitals, smartphone companies, or banks want to train machine learning models together, they face a dilemma: combining their data would produce better AI, but sharing sensitive patient records, personal messages, or financial information is often illegal or dangerous. Federated Learning, a privacy-preserving approach to machine learning, solves this by allowing organizations to train AI collaboratively without ever sharing their private data. This technology already powers voice assistants like Siri and Alexa. However, this privacy comes with a hidden cost. When something goes wrong, such as when an AI makes a mistake or produces harmful outputs, developers cannot investigate because they cannot see the original data. Traditional debugging and interpretability approaches, designed for centralized data access, are not directly applicable in this distributed, privacy-constrained setting. This thesis develops new methods that allow developers to identify which organization caused a problem and understand why the AI behaves as it does, all without ever accessing private data. We introduce techniques that work across different types of AI, from image recognition systems to advanced language models. These contributions, now integrated into widely-used federated learning platforms, make privacy-preserving AI more trustworthy and deployable in sensitive domains like healthcare and finance."],"dc:description.degree":["Doctor of Philosophy"],"dc:format.medium":["ETD"],"dc:identifier.other":["vt_gsexam:45318"],"dc:identifier.uri":["https://hdl.handle.net/10919/140660"],"dc:language.iso":["en"],"dc:publisher":["Virginia Tech"],"dc:rights":["In Copyright"],"dc:rights.uri":["http://rightsstatements.org/vocab/InC/1.0/"],"dc:subject":["Federated Learning","Privacy","Interpretability","Debugging","LLMs","CNNs"],"dc:title":["Interpretability and Debugging for Distributed Privacy Preserving Machine Learning"],"dc:type":["Dissertation"],"thesis:degree_discipline":["Computer Science & Applications"],"thesis:degree_level":["doctoral"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Virginia Polytechnic Institute and State University"]},"updated_at":"2026-07-22T22:20:03Z"}