Trust-Based Cloud Machine Learning Model Selection For Industrial IoT and Smart City Services

Basheer Qolomany, Ihab Mohammed, Ala I. Al-Fuqaha, Mohsen Guizani, Junaid Qadir

Introduction

The proliferation of the Industrial Internet of Things (IIoT) and Smart City services has led to the rise of Machine Learning as a Service (MLaaS). In these environments, cloud providers maintain large ensembles of machine learning (ML) models to process data from edge devices. However, deploying these models back to resource-constrained devices involves significant challenges regarding security, resource management, and computational complexity.

The proposed system architecture for trust-based model selection in MLaaS environments.

The Research Gap

Current research often treats resource constraints, adversarial security, and automated model selection in isolation. This paper addresses a critical research gap by integrating these domains into a unified framework that selects the most reliable models while respecting the operational limits of IoT hardware.

Venn diagram illustrating the research gap at the intersection of trust, security, and resource constraints.

Trusted Machine Learning (TML) Framework

The core contribution is the Trusted Machine Learning (TML) selection process. Trust is defined through consensus: a model is considered trustworthy if its predictions align with the majority of the ensemble. This "wisdom of the crowd" approach identifies models potentially compromised by "poisoning attacks," where malicious data is injected during training to degrade model integrity.

Optimization and Heuristics

Selecting the optimal sequence of models over time while minimizing reconfiguration overhead is an NP-complete problem. The authors first model this as an Integer Linear Programming (ILP) problem to establish a theoretical baseline. To make the system practical for real-time deployment, they introduce two polynomial-time heuristics:

  • Splice Heuristic: This identifies long, consecutive time segments where a single model maintains high trust, effectively "splicing" them together to minimize the need for frequent, energy-draining reconfigurations.
  • Fixing Heuristic: A more advanced approach that uses Linear Programming relaxation to find a solution nearly as accurate as the ILP but with significantly lower computational overhead.

A critical feature is the Fail-Safe Mechanism. If no model in the cloud pool meets a minimum trust threshold, the system triggers an alert, informing the device to enter a safe state rather than relying on potentially compromised data.

Experimental Evaluation

The researchers evaluated their approach using two primary proxies for real-world services: the CityPulse dataset for traffic prediction and the NASA Turbofan Engine Degradation dataset for predictive maintenance. To test the robustness of the trust metric, they intentionally poisoned 20% of the training data.

Key Results

  • Near-Optimal Performance: The proposed heuristics performed remarkably close to the "perfect" ILP solution. In the IIoT case study, the trust level was only 0.49% to 3.17% lower than the theoretical optimum.
  • Error Reduction: By selecting trusted models, the system achieved significantly lower Root Mean Square Error (RMSE) compared to using individual, potentially poisoned models—reducing error by up to 66.95% in traffic volume predictions.
  • Scalability: Unlike ILP, which struggles as the number of models or time slots increases, the heuristics provide a polynomial-time approximation, making them viable for large-scale cloud deployments.

Conclusion and Future Work

By leveraging ensemble consensus as a proxy for trust, this framework provides a robust defense against poisoning attacks while optimizing for the strict resource budgets of IoT devices. The authors suggest that future iterations could explore a wider variety of model architectures, such as CNNs, and address other adversarial threats like inference-time evasion attacks.

Summary updated

BibTeX
@article{qolomany2021trust,
  author = {Basheer Qolomany and Ihab Mohammed and Ala I. Al-Fuqaha and Mohsen Guizani and Junaid Qadir},
  title = {{Trust-Based Cloud Machine Learning Model Selection For Industrial IoT and Smart City Services}},
  journal = {IEEE Internet of Things Journal},
  year = {2021},
  doi = {10.1109/jiot.2020.3022323},
}
Figure