JODAR SOARES, Mr. Gustavo (2023) Implementation of secure environment with Intel SGX for image classification PRE - Research Project, ENSTA.
![]()
| PDF 1018Kb | |
![]()
| PDF 162Kb |
Abstract
This research internship focuses on developing a secure image classification system using artificial intelligence and Trusted Execution Environments (TEEs). The aim is to protect consumer personal image data while enabling its use for computational processes and statistical measures, making targeted advertisements based on image analysis possible, for example. The system follows a Personal Data Management System (PDMS) approach, ensuring automatic data collection and secure sharing of computed information. The architecture involves two data tasks, cmp and agg, where cmp uses a pre-trained model to classify images, and agg performs aggregation functions on the classified results. Intel SGX enclaves and the Artificial Intelligence library, TensorFlow Lite for Microcontrollers, were used to execute image classification tasks securely. Experiments were conducted using a simple neural network and the mobile_gallery_image_classification dataset and the results obtained demonstrate the effectiveness of the Replay technique in reducing classification time while maintaining high security levels, offering an efficient solution for personalized services across domains.
Item Type: | Thesis (PRE - Research Project) |
---|---|
Uncontrolled Keywords: | Personal data protection; Intel SGX; Image classification |
Subjects: | Information and Communication Sciences and Technologies |
ID Code: | 9591 |
Deposited By: | Gustavo JODAR SOARES |
Deposited On: | 25 août 2023 14:37 |
Dernière modification: | 25 août 2023 14:37 |
Repository Staff Only: item control page