A Novel Obfuscation Method Based on Majority Logic for Preventing Unauthorized Access to Binary Deep Neural Networks DOI
Alireza Mohseni, Mohammad Hossein Moaiyeri,

Mohammad Javad Adel

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 23, 2025

Abstract The significant expansion of deep learning applications has necessitated safeguarding the neural network (DNN) model from potential unauthorized access, highlighting its importance as a valuable asset. This study proposes an innovative key-based algorithm-hardware co-design methodology to protect models access. proposed approach significantly reduces accuracy when incorrect key is used, thereby preventing users accessing design. significance and advancements binary networks (BNNs) in hardware implementation cutting-edge DNN have led us develop our for BNNs. However, technique can be broadly applied various designs implementing accelerators. protective increases efficiency more than similar solutions across different BNN architectures standard datasets. We validate design using post-layout simulations Cadence Virtuoso tool well-established TSMC 40nm CMOS technology. yields reductions 43%, 79%, 71% area, average power, weight modification energy per filter structures. Additionally, security circuit been analyzed evaluated against Boolean satisfiability-based attacks, structural reverse engineering, power-based side-channel attacks.

Language: Английский

DeepSeek or ChatGPT: Can brain‐computer interfaces/brain‐inspired computing achieve leapfrog development with large AI models? DOI Creative Commons
Long Bai, Shugeng Chen, Peng Wang

et al.

Brain‐X, Journal Year: 2025, Volume and Issue: 3(1)

Published: March 1, 2025

Language: Английский

Citations

1
Laser-Guided Ion Dynamics in a Dual-Mode Memristor for Bioinspired Neuronal and Synaptic Integration DOI
Keon Jae Lee, Yu Jin Jeong, Kyunghwan Kim

et al.

Published: April 9, 2025

Abstract Neuromorphic computing aims to replicate the parallel, adaptive nature of biological intelligence in electronic systems. Despite considerable advances memristor technology, material-encoded neurosynaptic bifunctionality has not been demonstrated. We introduce a laser-guided dual-mode that integrates both volatility for neuronal spiking and nonvolatility synaptic plasticity within single-phase material. By precisely modulating silver ion dynamics through XeCl excimer laser irradiation, we achieve local dynamic control memristive behavior without requiring heterogeneous device array or stacking. The tunability with optimal computational efficiency demonstrates reconfigurable reservoir positive feedback loop learning.

Language: Английский

Citations

0
A Novel Obfuscation Method Based on Majority Logic for Preventing Unauthorized Access to Binary Deep Neural Networks DOI
Alireza Mohseni, Mohammad Hossein Moaiyeri,

Mohammad Javad Adel

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 23, 2025

Abstract The significant expansion of deep learning applications has necessitated safeguarding the neural network (DNN) model from potential unauthorized access, highlighting its importance as a valuable asset. This study proposes an innovative key-based algorithm-hardware co-design methodology to protect models access. proposed approach significantly reduces accuracy when incorrect key is used, thereby preventing users accessing design. significance and advancements binary networks (BNNs) in hardware implementation cutting-edge DNN have led us develop our for BNNs. However, technique can be broadly applied various designs implementing accelerators. protective increases efficiency more than similar solutions across different BNN architectures standard datasets. We validate design using post-layout simulations Cadence Virtuoso tool well-established TSMC 40nm CMOS technology. yields reductions 43%, 79%, 71% area, average power, weight modification energy per filter structures. Additionally, security circuit been analyzed evaluated against Boolean satisfiability-based attacks, structural reverse engineering, power-based side-channel attacks.

Language: Английский

Citations

0