{"id":{"repo_id":"vt","oai_identifier":"oai:vtechworks.lib.vt.edu:10919/139938"},"canonical_url":"https://search.dev.ndltd.org/etd/vt/oai:vtechworks.lib.vt.edu:10919/139938","repository":{"repo_id":"vt","name":"Virginia Tech","base_url":"https://vtechworks.lib.vt.edu/oai/request"},"display":{"title":"Efficient Distributed and Multi-Modal Machine Learning in Wireless Networks","abstract":"Next-generation wireless systems are expected to be artificial intelligence (AI)-native, in that they will embed machine learning (ML) and AI techniques from the application layer down to the physical layer. However, training and deploying ML models in wireless networks presents two key challenges pertaining to the limited computing and resources of wireless devices and systems, and the scarce and private nature of wireless data. First, ML models at the application layer, e.g., on-device AI, often require private data from distributed devices. One can resort to distributed ML algorithms such as federated learning (FL) by communicating only ML model parameters over wireless networks without sharing raw data. However, devices and communication networks have limited resources, in terms of computing, energy, bandwidth, and memory to support complex distributed ML algorithms. Second, next-generation wireless networks can potentially leverage a broad range of sensing modalities, such as LiDAR, images, or GPS, to make situation-aware network decisions under dynamic environments. To perform such wireless multi-modal data fusion, it is natural to leverage ML models and frameworks. However, the amount of training data is often scarce in wireless networks. As such, trained ML models often fail to generalize under unseen wireless environments. Moreover, despite providing more information about the current environment, multi-modal data also increases the amount of input features that must be processed by ML models. As such, multi-modal ML can suffer from large inference latency, thereby making obsolete network decisions in rapidly changing communication environments. The main contribution of this dissertation is, thus, to address these challenges by developing efficient and distributed ML frameworks that can be deployed over resource-constrained wireless networks with private and scarce multi-modal data. From the perspective of distributed, resource-efficient ML models, this dissertation first investigates energy-efficient distributed ML algorithms that operate over realistic wireless networks through the co-design of computing, communication, and learning algorithms. In particular, a novel energy-efficient FL framework is proposed to reduce the energy cost of training and communication by quantizing the neural network weights and activations. In this framework, every device trains a quantized neural network, which quantizes weights and activations to a limited precision level. The results show that the proposed framework can reduce energy consumption without jeopardizing the convergence rate by up to 70% compared to a baseline FL algorithm that does not use quantization. Subsequently, to optimize sparse model structures with low computational overhead, SpaFL: a communication-efficient FL framework is proposed. In SpaFL, a trainable threshold is defined for each filter/neuron to prune its all connected parameters, thereby leading to structured sparsity. The results showcase that SpaFL improves accuracy while requiring much less communication and computing resources compared to sparse baselines. This dissertation then investigates a large language model selection framework to optimize cost, latency, and response quality over 5G networks. In particular, a measurement-driven training framework is proposed for an AI-enabled router on a mobile device. The results show that the proposed framework can improve the cost and latency significantly with minimal response quality loss. From the perspective of efficient ML models, this dissertation first designs, efficient multi-modal learning frameworks are designed for improving ML generalization with scarce data and inference latency in wireless networks. In particular, a novel and data-efficient two-phase learning framework is proposed to improve generalization in unseen and unfamiliar wireless environments with a minimal amount of multi-modal data. In the first stage, a physics-based loss function is employed to enable each base station (BS) to learn the physics underlying its wireless environment captured by multi-modal data. In the second stage, collaborative domain adaptation is proposed to leverage the wireless environment knowledge of multiple BSs to guide under-performing BSs under domain shift. The results showcase that the proposed frameworks require significantly smaller amount of data and computing resources to achieve the convergence with better generalization. Next, a novel continual learning (CL) framework is proposed to achieve robust generalization to dynamic environments while retaining past knowledge. To this end, an agent estimates the distribution of risks over environmental change so as to obtain predictors that are robust to unseen changes. The results show that the proposed algorithm outperforms traditional CL baselines across all environments while significantly improving the generalization performance on unseen target environments. Lastly, a fast multi-modal transformer inference framework is designed to practically support wireless communication tasks by processing only important tokens. To validate the feasibility of the proposed framework for real-world deployments, one of the first multi-modal handover dataset is developed using a real-world testbed. The results show that the proposed framework can improve the inference latency by 86% compared to baselines with negligible performance loss. Overall, this dissertation develops a suite of efficient distributed and multi-modal ML frameworks that can be deployed in practical, real-world, and resource-constrained wireless networks.","abstract_html":"Next-generation wireless systems are expected to be artificial intelligence (AI)-native, in that they will embed machine learning (ML) and AI techniques from the application layer down to the physical layer. However, training and deploying ML models in wireless networks presents two key challenges pertaining to the limited computing and resources of wireless devices and systems, and the scarce and private nature of wireless data. First, ML models at the application layer, e.g., on-device AI, often require private data from distributed devices. One can resort to distributed ML algorithms such as federated learning (FL) by communicating only ML model parameters over wireless networks without sharing raw data. However, devices and communication networks have limited resources, in terms of computing, energy, bandwidth, and memory to support complex distributed ML algorithms. Second, next-generation wireless networks can potentially leverage a broad range of sensing modalities, such as LiDAR, images, or GPS, to make situation-aware network decisions under dynamic environments. To perform such wireless multi-modal data fusion, it is natural to leverage ML models and frameworks. However, the amount of training data is often scarce in wireless networks. As such, trained ML models often fail to generalize under unseen wireless environments. Moreover, despite providing more information about the current environment, multi-modal data also increases the amount of input features that must be processed by ML models. As such, multi-modal ML can suffer from large inference latency, thereby making obsolete network decisions in rapidly changing communication environments. The main contribution of this dissertation is, thus, to address these challenges by developing efficient and distributed ML frameworks that can be deployed over resource-constrained wireless networks with private and scarce multi-modal data. From the perspective of distributed, resource-efficient ML models, this dissertation first investigates energy-efficient distributed ML algorithms that operate over realistic wireless networks through the co-design of computing, communication, and learning algorithms. In particular, a novel energy-efficient FL framework is proposed to reduce the energy cost of training and communication by quantizing the neural network weights and activations. In this framework, every device trains a quantized neural network, which quantizes weights and activations to a limited precision level. The results show that the proposed framework can reduce energy consumption without jeopardizing the convergence rate by up to 70% compared to a baseline FL algorithm that does not use quantization. Subsequently, to optimize sparse model structures with low computational overhead, SpaFL: a communication-efficient FL framework is proposed. In SpaFL, a trainable threshold is defined for each filter/neuron to prune its all connected parameters, thereby leading to structured sparsity. The results showcase that SpaFL improves accuracy while requiring much less communication and computing resources compared to sparse baselines. This dissertation then investigates a large language model selection framework to optimize cost, latency, and response quality over 5G networks. In particular, a measurement-driven training framework is proposed for an AI-enabled router on a mobile device. The results show that the proposed framework can improve the cost and latency significantly with minimal response quality loss. From the perspective of efficient ML models, this dissertation first designs, efficient multi-modal learning frameworks are designed for improving ML generalization with scarce data and inference latency in wireless networks. In particular, a novel and data-efficient two-phase learning framework is proposed to improve generalization in unseen and unfamiliar wireless environments with a minimal amount of multi-modal data. In the first stage, a physics-based loss function is employed to enable each base station (BS) to learn the physics underlying its wireless environment captured by multi-modal data. In the second stage, collaborative domain adaptation is proposed to leverage the wireless environment knowledge of multiple BSs to guide under-performing BSs under domain shift. The results showcase that the proposed frameworks require significantly smaller amount of data and computing resources to achieve the convergence with better generalization. Next, a novel continual learning (CL) framework is proposed to achieve robust generalization to dynamic environments while retaining past knowledge. To this end, an agent estimates the distribution of risks over environmental change so as to obtain predictors that are robust to unseen changes. The results show that the proposed algorithm outperforms traditional CL baselines across all environments while significantly improving the generalization performance on unseen target environments. Lastly, a fast multi-modal transformer inference framework is designed to practically support wireless communication tasks by processing only important tokens. To validate the feasibility of the proposed framework for real-world deployments, one of the first multi-modal handover dataset is developed using a real-world testbed. The results show that the proposed framework can improve the inference latency by 86% compared to baselines with negligible performance loss. Overall, this dissertation develops a suite of efficient distributed and multi-modal ML frameworks that can be deployed in practical, real-world, and resource-constrained wireless networks.","abstract_has_math":false,"creators":["Kim, Minsu"],"institution":"Virginia Tech","degree_name":"Doctor of Philosophy","degree_level":"doctoral","degree_discipline":"Electrical Engineering","degree_department":"Electrical Engineering","school":null,"contributors":[],"advisors":[],"committee_chairs":["Saad, Walid"],"committee_members":["Stilwell, Daniel J.","Yang, Yaling","Ramakrishnan, Narendran","Dhillon, Harpreet Singh"],"year":2025,"date_issued":"2025-12-16","date_published":"2025-12-16","updated_at":"2026-07-22T22:20:21Z","subjects":["Machine Learning","Distributed Learning","Energy-efficiency","Data-Efficiency","Multi-Modal"],"languages":["en"],"rights":["Creative Commons Attribution-NonCommercial 4.0 International"],"rights_urls":["http://creativecommons.org/licenses/by-nc/4.0/"],"identifier_entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:44819"],"render_values":[{"text":"vt_gsexam:44819","href":null,"code":true}]}]},"links":{"outbound_url":"https://hdl.handle.net/10919/139938","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.committeechair","label":"Committee Chair","values":["Saad, Walid"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Stilwell, Daniel J.","Yang, Yaling","Ramakrishnan, Narendran","Dhillon, Harpreet Singh"]},{"key":"dc:contributor.department","label":"Department","values":["Electrical Engineering"]},{"key":"dc:creator","label":"Author","values":["Kim, Minsu"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-12-17T09:01:04Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-12-17T09:01:04Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-12-16"]},{"key":"dc:publisher","label":"Institution","values":["Virginia Tech"]},{"key":"dc:type","label":"Dc Type","values":["Dissertation"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering"]},{"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":["Machine Learning","Distributed Learning","Energy-efficiency","Data-Efficiency","Multi-Modal"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["en"]},{"key":"dc:rights","label":"Dc Rights","values":["Creative Commons Attribution-NonCommercial 4.0 International"]},{"key":"dc:rights.uri","label":"Rights URI","values":["http://creativecommons.org/licenses/by-nc/4.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.other","label":"Dc Identifier Other","values":["vt_gsexam:44819"]},{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10919/139938"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Next-generation wireless systems are expected to be artificial intelligence (AI)-native, in that they will embed machine learning (ML) and AI techniques from the application layer down to the physical layer. However, training and deploying ML models in wireless networks presents two key challenges pertaining to the limited computing and resources of wireless devices and systems, and the scarce and private nature of wireless data. First, ML models at the application layer, e.g., on-device AI, often require private data from distributed devices. One can resort to distributed ML algorithms such as federated learning (FL) by communicating only ML model parameters over wireless networks without sharing raw data. However, devices and communication networks have limited resources, in terms of computing, energy, bandwidth, and memory to support complex distributed ML algorithms. Second, next-generation wireless networks can potentially leverage a broad range of sensing modalities, such as LiDAR, images, or GPS, to make situation-aware network decisions under dynamic environments. To perform such wireless multi-modal data fusion, it is natural to leverage ML models and frameworks. However, the amount of training data is often scarce in wireless networks. As such, trained ML models often fail to generalize under unseen wireless environments. Moreover, despite providing more information about the current environment, multi-modal data also increases the amount of input features that must be processed by ML models. As such, multi-modal ML can suffer from large inference latency, thereby making obsolete network decisions in rapidly changing communication environments. The main contribution of this dissertation is, thus, to address these challenges by developing efficient and distributed ML frameworks that can be deployed over resource-constrained wireless networks with private and scarce multi-modal data. From the perspective of distributed, resource-efficient ML models, this dissertation first investigates energy-efficient distributed ML algorithms that operate over realistic wireless networks through the co-design of computing, communication, and learning algorithms. In particular, a novel energy-efficient FL framework is proposed to reduce the energy cost of training and communication by quantizing the neural network weights and activations. In this framework, every device trains a quantized neural network, which quantizes weights and activations to a limited precision level. The results show that the proposed framework can reduce energy consumption without jeopardizing the convergence rate by up to 70% compared to a baseline FL algorithm that does not use quantization. Subsequently, to optimize sparse model structures with low computational overhead, SpaFL: a communication-efficient FL framework is proposed. In SpaFL, a trainable threshold is defined for each filter/neuron to prune its all connected parameters, thereby leading to structured sparsity. The results showcase that SpaFL improves accuracy while requiring much less communication and computing resources compared to sparse baselines. This dissertation then investigates a large language model selection framework to optimize cost, latency, and response quality over 5G networks. In particular, a measurement-driven training framework is proposed for an AI-enabled router on a mobile device. The results show that the proposed framework can improve the cost and latency significantly with minimal response quality loss. From the perspective of efficient ML models, this dissertation first designs, efficient multi-modal learning frameworks are designed for improving ML generalization with scarce data and inference latency in wireless networks. In particular, a novel and data-efficient two-phase learning framework is proposed to improve generalization in unseen and unfamiliar wireless environments with a minimal amount of multi-modal data. In the first stage, a physics-based loss function is employed to enable each base station (BS) to learn the physics underlying its wireless environment captured by multi-modal data. In the second stage, collaborative domain adaptation is proposed to leverage the wireless environment knowledge of multiple BSs to guide under-performing BSs under domain shift. The results showcase that the proposed frameworks require significantly smaller amount of data and computing resources to achieve the convergence with better generalization. Next, a novel continual learning (CL) framework is proposed to achieve robust generalization to dynamic environments while retaining past knowledge. To this end, an agent estimates the distribution of risks over environmental change so as to obtain predictors that are robust to unseen changes. The results show that the proposed algorithm outperforms traditional CL baselines across all environments while significantly improving the generalization performance on unseen target environments. Lastly, a fast multi-modal transformer inference framework is designed to practically support wireless communication tasks by processing only important tokens. To validate the feasibility of the proposed framework for real-world deployments, one of the first multi-modal handover dataset is developed using a real-world testbed. The results show that the proposed framework can improve the inference latency by 86% compared to baselines with negligible performance loss. Overall, this dissertation develops a suite of efficient distributed and multi-modal ML frameworks that can be deployed in practical, real-world, and resource-constrained wireless networks."]},{"key":"dc:description.abstractgeneral","label":"General Abstract","values":["Next-generation wireless networks (like the sixth generation (6G) of wireless systems) will use artificial intelligence (AI) on mobile phones and within the networks themselves. For instance, on-device AI (e.g., Alexa and Apple Intelligence) is already being used on our mobile phones. To provide better connectivity, next-generation wireless networks can use AI models that can observe their surrounding environments. However, training and deploying such AI models in wireless networks will be challenging for two reasons. Firstly, AI models need a significant amount of resources such as computing hardware and energy for training. Moreover, the advent of large language models like ChatGPT imposes significant resource requirements, in terms of energy and compute, on devices such as mobile phones. Secondly, training data are often private and scarce in wireless networks. Privacy concerns primarily stem from the reluctance of data owners to share their data with operators of wireless networks. This dissertation addresses these practical AI challenges by pursuing two research thrusts focused on distributed and efficient AI designs for wireless networks. In the first thrust, this dissertation develops new techniques to efficiently train AI models for on-device AI with multiple mobile devices without sharing users' private data. A key focus area here is reducing the size of AI models without jeopardizing their performance. In the second thrust, this dissertation focuses on the design of new solutions that can efficiently train and deploy AI models in dynamic wireless networks to provide seamless connectivity and better communication services in presence of scarce multi-modal datasets (e.g., from cameras LiDAR, radar). To this end, novel algorithms are proposed that can learn the wireless channel propagation physics of surrounding environments and can improve the decision-making speed by focusing computation on the most important data. In summary, this dissertation addresses the challenges of training and deploying AI models in real-world wireless networks thereby improving energy efficiency, computing needs, and privacy of emerging AI algorithms."]},{"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":["Efficient Distributed and Multi-Modal Machine Learning in Wireless Networks"]}]}],"canonical_facts":{"dc:contributor.committeechair":["Saad, Walid"],"dc:contributor.committeemember":["Stilwell, Daniel J.","Yang, Yaling","Ramakrishnan, Narendran","Dhillon, Harpreet Singh"],"dc:contributor.department":["Electrical Engineering"],"dc:creator":["Kim, Minsu"],"dc:date.accessioned":["2025-12-17T09:01:04Z"],"dc:date.available":["2025-12-17T09:01:04Z"],"dc:date.issued":["2025-12-16"],"dc:description.abstract":["Next-generation wireless systems are expected to be artificial intelligence (AI)-native, in that they will embed machine learning (ML) and AI techniques from the application layer down to the physical layer. However, training and deploying ML models in wireless networks presents two key challenges pertaining to the limited computing and resources of wireless devices and systems, and the scarce and private nature of wireless data. First, ML models at the application layer, e.g., on-device AI, often require private data from distributed devices. One can resort to distributed ML algorithms such as federated learning (FL) by communicating only ML model parameters over wireless networks without sharing raw data. However, devices and communication networks have limited resources, in terms of computing, energy, bandwidth, and memory to support complex distributed ML algorithms. Second, next-generation wireless networks can potentially leverage a broad range of sensing modalities, such as LiDAR, images, or GPS, to make situation-aware network decisions under dynamic environments. To perform such wireless multi-modal data fusion, it is natural to leverage ML models and frameworks. However, the amount of training data is often scarce in wireless networks. As such, trained ML models often fail to generalize under unseen wireless environments. Moreover, despite providing more information about the current environment, multi-modal data also increases the amount of input features that must be processed by ML models. As such, multi-modal ML can suffer from large inference latency, thereby making obsolete network decisions in rapidly changing communication environments. The main contribution of this dissertation is, thus, to address these challenges by developing efficient and distributed ML frameworks that can be deployed over resource-constrained wireless networks with private and scarce multi-modal data. From the perspective of distributed, resource-efficient ML models, this dissertation first investigates energy-efficient distributed ML algorithms that operate over realistic wireless networks through the co-design of computing, communication, and learning algorithms. In particular, a novel energy-efficient FL framework is proposed to reduce the energy cost of training and communication by quantizing the neural network weights and activations. In this framework, every device trains a quantized neural network, which quantizes weights and activations to a limited precision level. The results show that the proposed framework can reduce energy consumption without jeopardizing the convergence rate by up to 70% compared to a baseline FL algorithm that does not use quantization. Subsequently, to optimize sparse model structures with low computational overhead, SpaFL: a communication-efficient FL framework is proposed. In SpaFL, a trainable threshold is defined for each filter/neuron to prune its all connected parameters, thereby leading to structured sparsity. The results showcase that SpaFL improves accuracy while requiring much less communication and computing resources compared to sparse baselines. This dissertation then investigates a large language model selection framework to optimize cost, latency, and response quality over 5G networks. In particular, a measurement-driven training framework is proposed for an AI-enabled router on a mobile device. The results show that the proposed framework can improve the cost and latency significantly with minimal response quality loss. From the perspective of efficient ML models, this dissertation first designs, efficient multi-modal learning frameworks are designed for improving ML generalization with scarce data and inference latency in wireless networks. In particular, a novel and data-efficient two-phase learning framework is proposed to improve generalization in unseen and unfamiliar wireless environments with a minimal amount of multi-modal data. In the first stage, a physics-based loss function is employed to enable each base station (BS) to learn the physics underlying its wireless environment captured by multi-modal data. In the second stage, collaborative domain adaptation is proposed to leverage the wireless environment knowledge of multiple BSs to guide under-performing BSs under domain shift. The results showcase that the proposed frameworks require significantly smaller amount of data and computing resources to achieve the convergence with better generalization. Next, a novel continual learning (CL) framework is proposed to achieve robust generalization to dynamic environments while retaining past knowledge. To this end, an agent estimates the distribution of risks over environmental change so as to obtain predictors that are robust to unseen changes. The results show that the proposed algorithm outperforms traditional CL baselines across all environments while significantly improving the generalization performance on unseen target environments. Lastly, a fast multi-modal transformer inference framework is designed to practically support wireless communication tasks by processing only important tokens. To validate the feasibility of the proposed framework for real-world deployments, one of the first multi-modal handover dataset is developed using a real-world testbed. The results show that the proposed framework can improve the inference latency by 86% compared to baselines with negligible performance loss. Overall, this dissertation develops a suite of efficient distributed and multi-modal ML frameworks that can be deployed in practical, real-world, and resource-constrained wireless networks."],"dc:description.abstractgeneral":["Next-generation wireless networks (like the sixth generation (6G) of wireless systems) will use artificial intelligence (AI) on mobile phones and within the networks themselves. For instance, on-device AI (e.g., Alexa and Apple Intelligence) is already being used on our mobile phones. To provide better connectivity, next-generation wireless networks can use AI models that can observe their surrounding environments. However, training and deploying such AI models in wireless networks will be challenging for two reasons. Firstly, AI models need a significant amount of resources such as computing hardware and energy for training. Moreover, the advent of large language models like ChatGPT imposes significant resource requirements, in terms of energy and compute, on devices such as mobile phones. Secondly, training data are often private and scarce in wireless networks. Privacy concerns primarily stem from the reluctance of data owners to share their data with operators of wireless networks. This dissertation addresses these practical AI challenges by pursuing two research thrusts focused on distributed and efficient AI designs for wireless networks. In the first thrust, this dissertation develops new techniques to efficiently train AI models for on-device AI with multiple mobile devices without sharing users' private data. A key focus area here is reducing the size of AI models without jeopardizing their performance. In the second thrust, this dissertation focuses on the design of new solutions that can efficiently train and deploy AI models in dynamic wireless networks to provide seamless connectivity and better communication services in presence of scarce multi-modal datasets (e.g., from cameras LiDAR, radar). To this end, novel algorithms are proposed that can learn the wireless channel propagation physics of surrounding environments and can improve the decision-making speed by focusing computation on the most important data. In summary, this dissertation addresses the challenges of training and deploying AI models in real-world wireless networks thereby improving energy efficiency, computing needs, and privacy of emerging AI algorithms."],"dc:description.degree":["Doctor of Philosophy"],"dc:format.medium":["ETD"],"dc:identifier.other":["vt_gsexam:44819"],"dc:identifier.uri":["https://hdl.handle.net/10919/139938"],"dc:language.iso":["en"],"dc:publisher":["Virginia Tech"],"dc:rights":["Creative Commons Attribution-NonCommercial 4.0 International"],"dc:rights.uri":["http://creativecommons.org/licenses/by-nc/4.0/"],"dc:subject":["Machine Learning","Distributed Learning","Energy-efficiency","Data-Efficiency","Multi-Modal"],"dc:title":["Efficient Distributed and Multi-Modal Machine Learning in Wireless Networks"],"dc:type":["Dissertation"],"thesis:degree_discipline":["Electrical Engineering"],"thesis:degree_level":["doctoral"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["Virginia Polytechnic Institute and State University"]},"updated_at":"2026-07-22T22:20:21Z"}