Machine learning and hardware security

Regazzoni, Francesco; Bhasin, Shivam; Pour, Amir Ali; Alshaer, Ihab; Aydin, Furkan; Aysu, Aydın; Beroulle, Vincent; Natale, Giorgio Di; Franzon, Paul D.; Hély, David; Homma, Naofumi; Ito, Akira; Jap, Dirmanto; Kashyap, Priyank; Polian, Ilia; Potluri, Seetal; Ueno, Rei; Vatajelu, Elena-Ioana; Yli-Mäyry, Ville

doi:10.1145/3400302.3416260

Cited by 13 publications

(3 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Due to the limited storage of TEE, Gangal et al [22] partitioned an edge device model and encapsulated only some layers in SGX powered TEE. However, the model parameters stored in unprotected memory are still easy to be stolen, and adversaries can build a complete model through model reverse engineering [23], [24]. Another idea is to explore building custom secure neural network accelerators [19], [25], [26].…”

Section: Hardware Protection Of Edge Device Modelsmentioning

confidence: 99%

Edge Deep Learning Model Protection via Neuron Authorization

Chen¹,

Zheng²,

Liu³

et al. 2023

Preprint

View full text Add to dashboard Cite

With the development of deep learning processors and accelerators, deep learning models have been widely deployed on edge devices as part of the Internet of Things. Edge device models are generally considered as valuable intellectual properties that are worth for careful protection. Unfortunately, these models have a great risk of being stolen or illegally copied. The existing model protections using encryption algorithms are suffered from high computation overhead which is not practical due to the limited computing capacity on edge devices. In this work, we propose a light-weight, practical, and general Edge device model Protection method at neuron level, denoted as EdgePro. Specifically, we select several neurons as authorization neurons and set their activation values to locking values and scale the neuron outputs as the "passwords" during training. EdgePro protects the model by ensuring it can only work correctly when the "passwords" are met, at the cost of encrypting and storing the information of the "passwords" instead of the whole model. Extensive experimental results indicate that EdgePro can work well on the task of protecting on datasets with different modes. The inference time increase of EdgePro is only 60% of state-of-the-art methods, and the accuracy loss is less than 1%. Additionally, EdgePro is robust against adaptive attacks including fine-tuning and pruning, which makes it more practical in real-world applications. EdgePro is also open sourced to facilitate future research: https://github.com/Leon022/EdgePro.

show abstract

Section: Hardware Protection Of Edge Device Modelsmentioning

confidence: 99%

Edge Deep Learning Model Protection via Neuron Authorization

Chen¹,

Zheng²,

Liu³

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The authors of the study [105] analyze the viability of repurposing an existing neural network to construct a robust Physically Unclonable Function in order to ensure safety and reliability in Internet of Things and smart sensor applications. The Multilayer Perceptron is the primary subject of this work.…”

Section: Machine Learning and Artificial Intelligencementioning

confidence: 99%

Building Trust in Microelectronics: A Comprehensive Review of Current Techniques and Adoption Challenges

Nyako,

Devkota,

et al. 2023

Electronics

View full text Add to dashboard Cite

The field of microelectronics has experienced extensive integration into various aspects of our everyday lives, evident via its utilization across a wide range of devices such as cellphones, airplanes, computers, wristwatches, and other similar technologies. Microelectronics are vital to the healthcare and defense industries, making them vulnerable to counterfeit products. Currently, the complicated global microelectronics supply chain involves the production of varied components in multiple places, resulting in tremendous risk. In this scenario, it is possible for hostile or adversarial actors to exploit the situation by intentionally introducing counterfeit components. This hostile behavior could steal data or use these components as remote kill switches. To address these problems, enormous resources are being committed to research, innovation, and development to build trust in microelectronics. This research study provides a thorough analysis of the taxonomy associated with prominent attack, detection, and avoidance models in the realm of counterfeit microelectronics. This research aims to improve our understanding of dependable microelectronics. Prevention strategies like Physical Unclonable Functions (PUFs) and machine learning (ML), and detection methods like aging-based fingerprints are reviewed in this study. Finally, we underscore the significance of interdisciplinary cooperation, commitment to norms, and proactive methods.

show abstract

“…In the contemporary digital era, information sharing has brought great convenience but also poses a tremendous risk of information leakage. Consequently, information security has become a significant focus in order to safeguard data security, leading to numerous advances in cryptographic technology. − Within this realm, metasurface-based optical encryption presents an excellent framework for the secure storage of encrypted information. − Optical encryption offers several unique merits, , including high speed, parallel processing, low power consumption, and numerous encoding channels. In tandem, the metasurface provides a compact subwavelength photonic platform with powerful light manipulation ability that can concurrently encode multiple optical parameters, − such as amplitude, , phase, , polarization, , and wavelength. , As a result, various metasurface-based optical encryption approaches are demonstrated to achieve high-security information storage based on distinct optical manipulation principles and encryption coding methods .…”

Section: Introductionmentioning

confidence: 99%

Compression-Encrypted Meta-Optics for Storage Efficiency and Security Enhancement

Wang,

Niu,

Zhao

et al. 2024

ACS Photonics

View full text Add to dashboard Cite

Information security is of vital importance in daily life, stimulating various cryptographic strategies to protect data from leaking. Among them, metasurface-based optical encryption is an excellent candidate because of its unique features, including numerous encoding channels and incomparable light-field manipulation ability. However, for state-of-the-art metasurface encryption, it is still critical to further improve its storage efficiency with high security for practical information-storage applications. Here, combining data compression and optical encryption, compression-encrypted meta-optics (CEM) is proposed to improve the regular metasurface storage efficiency by an order of magnitude while maintaining high information security. Utilizing lossless compression algorithms, the plaintexts are re-encoded into nonintuitive ciphertexts and keys with much less data volume (∼1/10), thus providing the first security barrier and increasing storage efficiency. Moreover, the polarization-encrypted metasurfaces also protect the ciphertexts from direct observation, providing a second security barrier. Overall, such a CEM with two security barriers significantly improves the optical storage efficiency and presents a novel route for potential applications in next-generation optical information storage/encryption.

show abstract

Machine learning and hardware security

Cited by 13 publications

References 14 publications

Edge Deep Learning Model Protection via Neuron Authorization

Edge Deep Learning Model Protection via Neuron Authorization

Building Trust in Microelectronics: A Comprehensive Review of Current Techniques and Adoption Challenges

Compression-Encrypted Meta-Optics for Storage Efficiency and Security Enhancement

Contact Info

Product

Resources

About