TT-SEAL: TTD-Aware Selective Encryption for Adversarially-Robust and Low-Latency Edge AI
#TT-SEAL #Tensor-Train Decomposition #Selective Encryption #Edge AI #Adversarial Robustness #Low-Latency #Model Compression #AES Encryption
📌 Key Takeaways
- TT-SEAL is a selective encryption framework specifically designed for TT-decomposed neural networks
- The framework ranks TT cores using a sensitivity-based importance metric
- TT-SEAL achieves robust security while encrypting only 4.89-15.92% of parameters
- The solution dramatically reduces AES decryption latency from 58% to 2.76% in tested models
- The research has been accepted at the Design Automation Conference 2026
📖 Full Retelling
A team of researchers led by Kyeongpil Min, including Sangmin Jeon, Jae-Jin Lee, and Woojoo Lee, developed TT-SEAL, a novel selective encryption framework for TT-decomposed networks, on February 24, 2026, to address the challenge of jointly satisfying model compression and security under tight device budgets in cloud-edge AI systems. The research addresses a critical gap in the field of edge AI security. While Tensor-Train Decomposition (TTD) effectively shrinks on-device models for deployment on resource-constrained edge devices, previous selective-encryption studies have largely assumed dense weights, leaving their practicality under TTD compression unclear. TT-SEAL introduces a novel approach that ranks TT cores with a sensitivity-based importance metric, calibrates a one-time robustness threshold, and uses a value-DP optimizer to encrypt only the minimum set of critical cores with AES encryption. Under TTD-aware, transfer-based threat models tested on an FPGA-prototyped edge processor, TT-SEAL demonstrates remarkable efficiency, matching the robustness of full (black-box) encryption while encrypting as little as 4.89-15.92% of parameters across various neural network architectures including ResNet-18, MobileNetV2, and VGG-16. More significantly, TT-SEAL reduces the share of AES decryption in end-to-end latency to low single digits (e.g., from 58% to 2.76% on ResNet-18), enabling secure, low-latency edge AI that was previously difficult to achieve with conventional approaches.
🏷️ Themes
Edge AI Security, Model Compression, Efficient Encryption
📚 Related People & Topics
Edge computing
Distributed computing paradigm
Edge computing is a distributed computing model that brings computation and data storage closer to the sources of data. More broadly, it refers to any design that pushes computation physically closer to a user, so as to reduce the latency compared to when an application runs on a centralized data ce...
Entity Intersection Graph
Connections for Edge computing:
🌐
Machine learning
1 shared
Original Source
--> Computer Science > Cryptography and Security arXiv:2602.22238 [Submitted on 24 Feb 2026] Title: TT-SEAL: TTD-Aware Selective Encryption for Adversarially-Robust and Low-Latency Edge AI Authors: Kyeongpil Min , Sangmin Jeon , Jae-Jin Lee , Woojoo Lee View a PDF of the paper titled TT-SEAL: TTD-Aware Selective Encryption for Adversarially-Robust and Low-Latency Edge AI, by Kyeongpil Min and 3 other authors View PDF HTML Abstract: Cloud-edge AI must jointly satisfy model compression and security under tight device budgets. While Tensor-Train Decomposition shrinks on-device models, prior selective-encryption studies largely assume dense weights, leaving its practicality under TTD compression unclear. We present TT-SEAL, a selective-encryption framework for TT-decomposed networks. TT-SEAL ranks TT cores with a sensitivity-based importance metric, calibrates a one-time robustness threshold, and uses a value-DP optimizer to encrypt the minimum set of critical cores with AES. Under TTD-aware, transfer-based threat models (and on an FPGA-prototyped edge processor) TT-SEAL matches the robustness of full (black-box) encryption while encrypting as little as 4.89-15.92% of parameters across ResNet-18, MobileNetV2, and VGG-16, and drives the share of AES decryption in end-to-end latency to low single digits (e.g., 58% -> 2.76% on ResNet-18), enabling secure, low-latency edge AI. Comments: 8 pages, 7 figures, 3 tables. This paper has been accepted at Design Automation Conference 2026 Subjects: Cryptography and Security (cs.CR) ; Artificial Intelligence (cs.AI) Cite as: arXiv:2602.22238 [cs.CR] (or arXiv:2602.22238v1 [cs.CR] for this version) https://doi.org/10.48550/arXiv.2602.22238 Focus to learn more arXiv-issued DOI via DataCite Submission history From: Woojoo Lee [ view email ] [v1] Tue, 24 Feb 2026 05:48:09 UTC (7,969 KB) Full-text links: Access Paper: View a PDF of the paper titled TT-SEAL: TTD-Aware Selective Encryption for Adversarially-Robust and Low-Latency Edge AI, ...
Read full article at source