Authors:
B. Vijayalakshmi, P. Jayalakshmi, B. Sarvesan, A. Vishnukumar, G. Ragu, S. Saranya, Rahul Panakkal
Addresses:
Department of General Engineering, R.M.K. College of Engineering and Technology, Thiruvallur, Tamil Nadu, India. Department of Computer Science and Engineering, SRM Institute of Science and Technology, Ramapuram, Chennai, Tamil Nadu, India. Department of Information Technology, R.M.D. Engineering College, Thiruvallur, Tamil Nadu, India. Department of Computer Science and Business Systems, Rajalakshmi Engineering College, Chennai, Tamil Nadu, India. Department of Computer Science and Engineering, Dhaanish Ahmed College of Engineering, Chennai, Tamil Nadu, India. Department of Computer Science and Engineering, University of Illinois at Urbana, Champaign, Illinois, United States of America.
Abstract:
Although researchers have only begun to scratch the surface, this paper focuses on the fundamental trade-off between data privacy and analytical utility in an edge-computing scenario. Classical privacy-preserving strategies suffer from. performance degradation in multimodal analysis when uniform noise is applied without considering the semantic context. To address this, researchers present a. Context-Aware Differential Privacy system that adapts the noise level depending on the sensitivity of the detected environment and its specific data modalities. Researchers also incorporate a provable utility guarantee to ensure that the privacy-preserving transformations do not render the data unusable for subsequent tasks. The dataset consists of 445 examples of multimodal sensor input (video frames and audio spectrograms). The framework was then implemented in Python; the libraries used included TensorFlow Privacy for noise injection and Edge Impulse for on-device inference simulation. Results show that our approach can maintain high classification accuracy with a tight privacy budget and significantly outperforms static differential privacy techniques. This work provides a strong foundation for deploying secure, high-value analytics on resource-constrained edge devices.
Keywords: Edge Computing; Differential Privacy; Multimodal Analytics; Verifiable Utility; Context Awareness; Classical Privacy; Edge-Computing Scenario; Tensor-Flow Privacy; Edge Impulse.
Received on: 20/04/2025, Revised on: 15/08/2025, Accepted on: 02/10/2025, Published on: 05/05/2026
DOI: 10.64091/ATICS.2026.000258
AVE Trends in Intelligent Computing Systems, 2026 Vol. 3 No. 2 , Pages: 86-94