慣性聚合 高效追讀感興趣之博客、新聞、科技資訊
閱原文 以慣性聚合開啟

推薦訂閱源

博客园 - 司徒正美
V
V2EX
T
Tailwind CSS Blog
有赞技术团队
有赞技术团队
aimingoo的专栏
aimingoo的专栏
Apple Machine Learning Research
Apple Machine Learning Research
IT之家
IT之家
Blog — PlanetScale
Blog — PlanetScale
A
About on SuperTechFans
月光博客
月光博客
T
The Blog of Author Tim Ferriss
宝玉的分享
宝玉的分享
Martin Fowler
Martin Fowler
博客园 - 聂微东
The GitHub Blog
The GitHub Blog
V
Visual Studio Blog
WordPress大学
WordPress大学
酷 壳 – CoolShell
酷 壳 – CoolShell
Engineering at Meta
Engineering at Meta
GbyAI
GbyAI

Cryptology ePrint Archive

Dynamic Group Time-based One-time Passwords MBU: Scalable and Constant-Round Evaluation of Non-linear Functions in Standard MPC Setting CEDAR: A Compact and Efficient Decoder Architecture for RS-RM Code in HQC LEAH: Lightweight and Efficient Hardware Accelerator for Code-based PQC Scheme HQC The Sum-Check Protocol over the Monomial Basis, and Other Optimizations How to Authenticate a Non-Deterministic Computation Fully Adaptive Threshold Blind Signature Without AGM Additive FFTs for HQC on ARM Cortex-M4, Revisited SPoCK: Sequential Proofs of Complete Knowledge Blind Verifiable Delay Functions Compact Fully Asynchronous Updatable Public Key Encryption Scheme from Hamming Quasi-Cyclic Cryptosystem CLAASP-MP: An Automated MILP Framework for Monomial Prediction Assessing Geometric Security of AES Neural Realizations: Linear-Time Key Recovery via Neural Leakage Explicit Bounds on the Existence Probability of Random Multivariate Quadratic Systems over Finite Fields Faster Logical Operations from Discrete CKKS SecDTD: Dynamic Token Drop for Secure Transformers Inference Towards Zero Rotation and Beyond: Architecting Neural Networks for Fast Secure Inference with Homomorphic Encryption Code-based Scalable Collaborative SNARKs MAGNET: MAsked Gaussian Now Efficient and Table-less Automated formal analysis of Signal’s Double Ratchet: attacks, fixes and security proofs Bitsliced Segment-Based Search Technique for Low-Depth and Hardware-Efficient S-Box Circuits The Cost of Fluidity: Communication Complexity Trade-offs in Fluid MPC Decomposition of Large Look-Up Tables for Fast Homomorphic Evaluation Hint-Free Multi-Signatures (Mis)using the Lattice Isomorphism Problem. Cryptanalysis of the double-LIP and Construction of LIP-Based Blind Signatures Improving LatticeFold+ with ℓ2-norm Checks ABRA-CAPA-DABRA: Full break of CAPA Polynomial-Time Cryptanalytic Extraction of Graph Neural Networks in the Hard-Label Setting Zero-Knowledge Proof of Progress: Secure Multi-Phase Capture-the-Flag Competitions Scalable Registration-Based Encryption from Lattices ISE-supported erasure of residual shares Chorus: Secret Recovery with Ephemeral Client Committees $\mathsf{Veloz}$: Efficient and Flexible Distribution Framework for Code-Based Polynomial Commitment Scheme How to construct even faster and indifferentiable hash functions from random permutations Public Key Encryption from High-Corruption Constraint Satisfaction Problems Drop-In Masked Modular Reduction for ML-DSA: Cutting Side-Channel Cost in the Root-of-Trust Optimizing and Implementing Threshold MAYO Improved Cryptanalysis of the Permuted Kernel Problem with Applications to PERK v2.2.0, SUSHSYFISH and PKP-DSS Cross-Paradigm Models of Restricted Syndrome Decoding with Application to CROSS Fast Isogeny Evaluation on Binary Curves Quick Draw Queries: Lightweight Searchable Public-key Ciphertexts with Hidden Structures via Non-Interactive Key Exchange A Constructive Treatment of Authentication Boolean Arithmetic over $\mathbb{F}_2$ from Group Commutators HAWK with Hint: Algebraic Key Recovery from Side-Channel Leakage Post-Quantum Secure k-Times Traceable Ring Signature A Key Schedule Design and Evaluation under Boundary Round-Key Leakage 2G2T: Constant-Size, Statistically Sound MSM Outsourcing Proximity Signatures Breaking Optimized HQC: The First Cache-Timing Full Decryption Oracle Key-Recovery Attack in Post-Quantum Cryptography Efficient Partially Blind Signatures from Isogenies
几何临界点筛选:无聚类之密码分析提取神经网络模型
Ming Duan · 2026-05-22 · via Cryptology ePrint Archive

论文 2026/1025

几何临界点筛选:无聚类之密码分析提取神经网络模型

唐沛瑶,信息工程学院,郑州,中国

摘要

神經網絡模型萃取,近年已成為關鍵安全議題。庚子年,卡拉利尼等將模型萃取分為簽名萃取與跡萃取。甲辰年,卡納勒斯-馬丁內斯等提出一多項式時間跡萃取之法。丙午年,劉氏等成首例八層深神經網絡模型萃取之功。然現有簽名萃取之法,遵循先計算後聚類之舊制:首計無名層巨量候選關鍵點之簽名,再以聚類分層,致查詢與計算之費過鉅。 是文呈現以幾何關係為基之關鍵點篩選之法。於三共面平行線上尋關鍵點,可疾速分離首隱藏層神經元之關鍵點,以最少簽名萃取,減簽名萃取之查詢複雜度自 $O(N \log N \cdot d_0)$ 至 $O(d_1\cdot d_0)$。於輸入維度逾首隱藏層維度之神經網絡,更可於全激活空間內三共面線段上尋關鍵點,以疾速篩選二隱藏層關鍵點,而首隱藏層神經元皆已激活。 幾何關鍵點篩選,僅需計算少量非目標關鍵點之簽名。其優在查詢之費低,且能自動驗證汙染之關鍵點。於 $784-8^{(8)}-1$ 網絡實驗,首二隱藏層神經元簽名萃取所需之時,僅為現有之法百分之一點七與百分之三點七,查詢之費減至最前沿之法百分之三點二與百分之零點一。復次,此法不限於 ReLU 激活,可推廣至其他分段線性激活函數,為神經網絡模型萃取提供根本且通輕之法。

BibTeX

@misc{cryptoeprint:2026/1025,
      author = {Ming Duan and Peiyao Tang},
      title = {Geometric Critical Point Screening: Clustering-Free Cryptanalytic Extraction of Neural Network Models},
      howpublished = {Cryptology {ePrint} Archive, Paper 2026/1025},
      year = {2026},
      url = {https://eprint.iacr.org/2026/1025}
}