← All research areas

Secure & Sustainable AI Systems

Understanding the joint costs of intelligence, robustness and resource use across learning and deployment.

Conceptual illustration of secure & sustainable ai systems

Research Description

Overview

We study security and efficiency together, asking how attacks, defenses and learning strategies change predictive utility, energy use and response time. This matters when battery budgets, response deadlines and computing costs limit deployment of on-device AI, LLMs and agentic services across edge and cloud systems.

Existing work examines energy–latency attacks, model-pruning defenses, environmental-cost measurement and robustness–sustainability tradeoffs. Resource-aware agents and data-center-scale evaluation are emerging directions, not established infrastructure deployments.

Research Thrusts

Resource-aware resilience

Measure robustness alongside computational cost, memory, latency and energy rather than optimizing accuracy in isolation.

Efficient learning and deployment

Investigate how model choices and deployment constraints interact in edge and resource-constrained AI systems.

Emerging AI infrastructure

Extend questions about efficiency and security toward agent systems and AI infrastructure. Data-center-scale research remains a proposed direction, not an established deployment.

Research Goals

Measure predictive utility, adversarial resilience, energy and latency together, including resource-constrained settings.

Intellectual Merit

Make security–sustainability tradeoffs explicit and comparable across methods and system conditions.

Research Impact

The intended contribution is evidence for choosing AI designs under practical resource and security constraints. Benefits must be measured for each workload and platform rather than assumed from a smaller model alone.

Current Projects & Emerging Directions

Published and preprint foundations

Resource-aware adversarial resilience

Sensing workloadAttack / defenseAccuracy + energy + latency

Attacks can change more than a prediction: they can also increase the time and energy required to process sensor data. This project examines that broader attack surface and the resource costs introduced by defenses.

Research objectives

Evaluate predictive performance alongside energy and latency, including the tradeoffs created by model pruning.

Approach & Intellectual Merit

The published Sponge Attacks study provides a sensing-AI foundation. Comparisons should account for model, device and workload so that apparent efficiency gains are not confused with reduced utility.

Selected publications

Sponge attacks on sensing ai: Energy-latency vulnerabilities and defense via model pruning ↗

Relevance & Applications

Resource-constrained sensing and edge platforms need dependable response times as well as accurate outputs. These evaluations help identify when a defense shifts costs elsewhere rather than removing them.

Impact

The linked work studies energy–latency vulnerabilities and model-pruning defenses. Its potential benefit is a fuller basis for comparing security costs; this is not a claim of deployed protection or measured real-world savings.

Existing research

Sustainable machine learning

Training & defenseMeasure resource useExplain tradeoffs

Security measures are part of the computational workload of AI. Training, repeated evaluation and adversarial defenses can change energy requirements and their associated environmental costs.

Research objectives

Make resource and environmental measurements transparent and compare the costs of learning and defense under clearly stated assumptions.

Approach & Intellectual Merit

Published work reviews carbon-emission quantification and examines the sustainability and social costs of adversarial training. The approach emphasizes the measurement boundary, hardware and energy assumptions behind an estimate.

Selected publications

Selected sustainability publications ↗

Relevance & Applications

Researchers can use these distinctions to report security–efficiency tradeoffs more responsibly. Estimated carbon costs are not equivalent to measured reductions in operational emissions.

Impact

Published work reviews carbon-emission quantification and examines the costs of adversarial training. The intended benefit is more transparent tradeoff reporting, not a claim of verified emissions reductions in deployed systems.

Emerging direction · not an established deployment

AI infrastructure and agent systems

Devices & agentsEdge / cloud coordinationSecurity + resource budgets

As AI workloads move between devices, edge services and larger infrastructure, security and resource decisions become interconnected. Agent workflows introduce additional questions about the information and computation passed between components.

Research objectives

Develop an evaluation agenda for trustworthy coordination, resource allocation and evidence flow across connected AI systems.

Approach & Intellectual Merit

This is a proposed systems direction informed by existing efficiency and security research. Data-center-scale evaluation is a future target; no dedicated infrastructure deployment or project-specific performance result is asserted.

Relevance & Applications

The long-term relevance is to resource-aware AI services whose security cannot be assessed at the model level alone. Operational impact will require system-scale experiments and independent validation.

Impact

The intended impact is a framework for studying how security, resource constraints, and evidence flow interact at system scale. This remains an emerging agenda; no project-specific outcomes, infrastructure deployment, or operational savings are claimed.

Related Publications

JIF 5.32025 metricJournal

Quantifying Robustness and Sustainability Trade-off in Federated Adversarial Learning for Cyber-Physical Systems

Syed Mhamudul Hasan, Ahmed Imteaj, Abdur R Shahid

IEEE Transactions on Sustainable Computing · 2026

AI Security & PrivacyFederated LearningSustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@article{shield70,
  title = {Quantifying Robustness and Sustainability Trade-off in Federated Adversarial Learning for Cyber-Physical Systems},
  author = {Syed Mhamudul Hasan and Ahmed Imteaj and Abdur R Shahid},
  year = {2026},
  journal = {IEEE Transactions on Sustainable Computing},
  publisher = {IEEE},
  doi = {10.1109/TSUSC.2025.3633995},
  url = {https://doi.org/10.1109/TSUSC.2025.3633995}
}
Symposium

Benchmarking large-language models for resource-efficient medical AI for edge deployment

Awal Ahmed Fime, Md Zarif Hossain, Saika Zaman, Abdur Rahman Bin Shahid, Ahmed Imteaj

Proceedings of the AAAI Symposium Series · vol. 5, no. 1, pp. 137-142, 2025

LLMsSustainable & Resource-Efficient AIHealthcare AI
Summary
This paper compares parameter-efficient fine-tuning of large language models for medical question answering, considering answer quality and resource efficiency for edge deployment.
PDFPaper ↗DOI ↗
BibTeX
@inproceedings{shield59,
  title = {Benchmarking large-language models for resource-efficient medical AI for edge deployment},
  author = {Awal Ahmed Fime and Md Zarif Hossain and Saika Zaman and Abdur Rahman Bin Shahid and Ahmed Imteaj},
  year = {2025},
  booktitle = {Proceedings of the AAAI Symposium Series},
  volume = {5},
  number = {1},
  pages = {137-142},
  doi = {10.1609/aaaiss.v5i1.35580},
  url = {https://doi.org/10.1609/aaaiss.v5i1.35580}
}
JIF 5.32025 metricJournal

Carbon emission quantification of machine learning: A review

Syed Mhamudul Hasan, Taminul Islam, Munshi Saifuzzaman, Khaled R. Ahmed, Chun-Hsi Huang, Abdur R. Shahid

IEEE Transactions on Sustainable Computing · vol. 10, no. 6, pp. 1085-1102, 2025

Sustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@article{shield64,
  title = {Carbon emission quantification of machine learning: A review},
  author = {Syed Mhamudul Hasan and Taminul Islam and Munshi Saifuzzaman and Khaled R. Ahmed and Chun-Hsi Huang and Abdur R. Shahid},
  year = {2025},
  journal = {IEEE Transactions on Sustainable Computing},
  volume = {10},
  number = {6},
  pages = {1085-1102},
  publisher = {IEEE},
  doi = {10.1109/TSUSC.2025.3578834},
  url = {https://doi.org/10.1109/TSUSC.2025.3578834}
}
JIF 5.02025 metricJournal

Evaluating sustainability and social costs of adversarial training in machine learning

Syed Mhamudul Hasan, Abdur R. Shahid, Ahmed Imteaj

IEEE Consumer Electronics Magazine · vol. 14, no. 3, pp. 27-34, 2025

AI Security & PrivacySustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@article{shield51,
  title = {Evaluating sustainability and social costs of adversarial training in machine learning},
  author = {Syed Mhamudul Hasan and Abdur R. Shahid and Ahmed Imteaj},
  year = {2025},
  journal = {IEEE Consumer Electronics Magazine},
  volume = {14},
  number = {3},
  pages = {27-34},
  publisher = {IEEE},
  doi = {10.1109/MCE.2024.3458350},
  url = {https://doi.org/10.1109/MCE.2024.3458350}
}
IEEE GLOBECOMConference

Sponge attacks on sensing ai: Energy-latency vulnerabilities and defense via model pruning

Syed Mhamudul Hasan, Hussein Zangoti, Iraklis Anagnostopoulos, Abdur R. Shahid

GLOBECOM 2025-2025 IEEE Global Communications Conference · pp. 4445-4450, 2025

AI Security & PrivacySustainable & Resource-Efficient AIWearable Sensing
Paper ↗DOI ↗
BibTeX
@inproceedings{shield57,
  title = {Sponge attacks on sensing ai: Energy-latency vulnerabilities and defense via model pruning},
  author = {Syed Mhamudul Hasan and Hussein Zangoti and Iraklis Anagnostopoulos and Abdur R. Shahid},
  year = {2025},
  booktitle = {GLOBECOM 2025-2025 IEEE Global Communications Conference},
  pages = {4445-4450},
  publisher = {IEEE},
  doi = {10.1109/GLOBECOM59602.2025.11432163},
  url = {https://doi.org/10.1109/GLOBECOM59602.2025.11432163}
}
Conference

Distributed threat intelligence at the edge devices: A large language model-driven approach

Syed Mhamudul Hasan, Alaa M Alotaibi, Sajedul Talukder, Abdur R Shahid

2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC) · pp. 1496-1497, 2024

AI Security & PrivacyLLMs
Paper ↗DOI ↗
BibTeX
@inproceedings{shield40,
  title = {Distributed threat intelligence at the edge devices: A large language model-driven approach},
  author = {Syed Mhamudul Hasan and Alaa M Alotaibi and Sajedul Talukder and Abdur R Shahid},
  year = {2024},
  booktitle = {2024 IEEE 48th Annual Computers, Software, and Applications Conference (COMPSAC)},
  pages = {1496-1497},
  publisher = {IEEE},
  doi = {10.1109/compsac61105.2024.00206},
  url = {https://doi.org/10.1109/compsac61105.2024.00206}
}
Conference

Enhancing Road Safety through Cost-Effective, Real-Time Monitoring of Driver Awareness with Resource-Constrained IoT Devices

Ahmed Imteaj, Tanveer Rahman, Saika Zaman, Md Zarif Hossain, Abdur R. Shahid

48th IEEE International Conference on Computers, Software, and Applications (COMPSAC 2024) · 2024

Cyber-Physical Systems & IoTSustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@inproceedings{shield44,
  title = {Enhancing Road Safety through Cost-Effective, Real-Time Monitoring of Driver Awareness with Resource-Constrained IoT Devices},
  author = {Ahmed Imteaj and Tanveer Rahman and Saika Zaman and Md Zarif Hossain and Abdur R. Shahid},
  year = {2024},
  booktitle = {48th IEEE International Conference on Computers, Software, and Applications (COMPSAC 2024)},
  doi = {10.1109/compsac61105.2024.00270},
  url = {https://doi.org/10.1109/compsac61105.2024.00270}
}
Conference

Fedavo: Improving communication efficiency in federated learning with african vultures optimizer

Md Zarif Hossain, Ahmed Imteaj, Abdur R. Shahid

Accepted in 48th IEEE International Conference on Computers, Software, and Applications (COMPSAC 2024) arXiv preprint arXiv:2305.01154 · 2024

Federated LearningSustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@inproceedings{shield43,
  title = {Fedavo: Improving communication efficiency in federated learning with african vultures optimizer},
  author = {Md Zarif Hossain and Ahmed Imteaj and Abdur R. Shahid},
  year = {2024},
  booktitle = {Accepted in 48th IEEE International Conference on Computers, Software, and Applications (COMPSAC 2024) arXiv preprint arXiv:2305.01154},
  doi = {10.1109/compsac61105.2024.00069},
  url = {https://doi.org/10.1109/compsac61105.2024.00069}
}
Workshop

FLAMINGO: Adaptive and Resilient Federated Meta-Learning against Adversarial Attacks

Md Zarif Hossain, Ahmed Imteaj, Abdur R. Shahid

2024 IEEE 44th International Conference on Distributed Computing Systems Workshops (ICDCSW) · 2024

AI Security & PrivacyFederated Learning
PDFPaper ↗DOI ↗
BibTeX
@inproceedings{shield42,
  title = {FLAMINGO: Adaptive and Resilient Federated Meta-Learning against Adversarial Attacks},
  author = {Md Zarif Hossain and Ahmed Imteaj and Abdur R. Shahid},
  year = {2024},
  booktitle = {2024 IEEE 44th International Conference on Distributed Computing Systems Workshops (ICDCSW)},
  doi = {10.1109/icdcsw63686.2024.00009},
  url = {https://doi.org/10.1109/ICDCSW63686.2024.00009}
}
Conference

The Environmental Price of Intelligence: Evaluating the Social Cost of Carbon in Machine Learning

Syed Mhamudul Hasan, Abdur R. Shahid, Ahmed Imteaj

11th IEEE Conference on Technologies for Sustainability (SusTech 2024) · 2024

Sustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@inproceedings{shield33,
  title = {The Environmental Price of Intelligence: Evaluating the Social Cost of Carbon in Machine Learning},
  author = {Syed Mhamudul Hasan and Abdur R. Shahid and Ahmed Imteaj},
  year = {2024},
  booktitle = {11th IEEE Conference on Technologies for Sustainability (SusTech 2024)},
  doi = {10.1109/sustech60925.2024.10553496},
  url = {https://doi.org/10.1109/sustech60925.2024.10553496}
}
IEEE ICC WorkshopWorkshop

Towards sustainable secureml: Quantifying carbon footprint of adversarial machine learning

Syed Mhamudul Hasan, Abdur R. Shahid, Ahmed Imteaj

2024 IEEE International Conference on Communications Workshops (ICC Workshops) · pp. 1359-1364, 2024

AI Security & PrivacySustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@inproceedings{shield39,
  title = {Towards sustainable secureml: Quantifying carbon footprint of adversarial machine learning},
  author = {Syed Mhamudul Hasan and Abdur R. Shahid and Ahmed Imteaj},
  year = {2024},
  booktitle = {2024 IEEE International Conference on Communications Workshops (ICC Workshops)},
  pages = {1359-1364},
  publisher = {IEEE},
  doi = {10.1109/ICCWorkshops59551.2024.10615723},
  url = {https://doi.org/10.1109/ICCWorkshops59551.2024.10615723}
}
Conference

Tripleplay: Enhancing federated learning with clip for non-iid data and resource efficiency

Ahmed Imteaj, Md Zarif Hossain, Saika Zaman, Abdur R Shahid

2024 International Conference on Machine Learning and Applications (ICMLA) · pp. 1474-1480, 2024

Federated LearningMultimodal AISustainable & Resource-Efficient AI
Paper ↗DOI ↗
BibTeX
@inproceedings{shield50,
  title = {Tripleplay: Enhancing federated learning with clip for non-iid data and resource efficiency},
  author = {Ahmed Imteaj and Md Zarif Hossain and Saika Zaman and Abdur R Shahid},
  year = {2024},
  booktitle = {2024 International Conference on Machine Learning and Applications (ICMLA)},
  pages = {1474-1480},
  publisher = {IEEE},
  doi = {10.1109/ICMLA61862.2024.00228},
  url = {https://doi.org/10.1109/ICMLA61862.2024.00228}
}