安全多方计算（MPC）的威胁模型假设了什么？
What does the threat model of secure multi-party computation (MPC) assume?
MPC 假设参与方互不信任，但其中最多有 t 个（可设）会作恶或串谋；它保证即使这些恶意方合谋，也学不到其他方的私输入，只能得到约定的计算结果。它不抵抗所有方一起作弊。
MPC assumes mutually distrusting parties where at most t (a chosen threshold) may cheat or collude; it guarantees they learn nothing beyond the agreed result even if they collude, but cannot stop everyone cheating together.
难度 专业 · 适合 大众
CC BY 4.0 · 已核验 · Verified 2026-11-03
来源 · Source
Goldwasser、Micali 与 Rackoff《交互式证明系统的知识复杂度》(SIAM Journal on Computing, 1985)；Katz 与 Lindell《现代密码学引论》
Goldwasser, Micali & Rackoff, 'The Knowledge Complexity of Interactive Proof Systems' (SIAM Journal on Computing, 1985); Katz & Lindell, 'Introduction to Modern Cryptography'
独立核验依据 · Independent verification: 核验依据 1 OPENLIBRARY 标识 /works/OL508869W：Introduction to modern c
#MPC #安全
同态加密允许在密文上做哪些计算？
What computations does homomorphic encryption allow on ciphertext?
同态加密让你可以直接对加密数据做加法、乘法等运算，得到的结果解密后等于对明文做同样运算的结果。这样云服务器能在看不到原始数据的情况下完成计算，典型应用如加密的投票统计或隐私查询。
Homomorphic encryption lets you add or multiply encrypted data so that decrypting the result equals doing the same op on plaintext. A server can compute without seeing the data—e.g. tallying encrypted votes or private queries.
难度 专业 · 适合 大众
CC BY 4.0 · 来源待核验 · Source pending review
来源 · Source
出处备注：此条未指向带版次的可引用著作（通识性内容、机构文件或未展开的英文书名），本库不提供独立书目出处；如需精确引用，需由编辑补全具体篇目与版次。
Gentry, A Fully Homomorphic Encryption Scheme, PhD Thesis, Stanford 2009
Gentry, A Fully Homomorphic Encryption Scheme, PhD Thesis, Stanford 2009
#同态加密 #隐私
联邦学习怎么做到不把本地数据传出去也能训练模型？
How does federated learning train without sending local data out?
各设备先用本地数据训练模型，只把更新后的参数或梯度（而非原始数据）加密上传，服务器聚合成一个更好的全局模型再下发。原始数据始终留在设备上，从而保护隐私。
Each device trains locally and uploads only model updates or gradients—not raw data—which the server aggregates into a better global model and sends back. Raw data never leaves the device.
难度 进阶 · 适合 大众
CC BY 4.0 · 已核验 · Verified 2026-11-04
来源 · Source
《Federated Learning: Collaborative Machine Learning Without Data Sharing》,Journal of Computational Analysis and Applications,2025 (DOI: 10.48047/jocaaa.2025.34.12.02)
'Federated Learning: Collaborative Machine Learning Without Data Sharing', Journal of Computational Analysis and Applications, 2025, DOI: 10.48047/jocaaa.2025.34.12.02
独立核验依据 · Independent verification: 核验依据 1 CROSSREF 标识 10.48047/jocaaa.2025.34.12.02：Federated Lear
#联邦学习 #隐私
差分隐私是如何用数学方式保护个人数据的？
How does differential privacy mathematically protect individuals?
差分隐私在统计结果里加入经过标定的随机噪声，使得任何单条记录“在或不在”数据集中都不会显著改变输出。这样即使攻击者拿到结果，也难以反推出某个具体人的信息，并用参数 ε 控制泄露程度。
Differential privacy adds carefully calibrated random noise so that any single record's presence barely changes the output. An attacker seeing the result can't pin down one person, with a parameter ε bounding the leakage.
难度 进阶 · 适合 大众
CC BY 4.0 · 已核验 · Verified 2026-11-04
来源 · Source
Dwork 与 Roth,《The Algorithmic Foundations of Differential Privacy》,Foundations and Trends® in Theoretical Computer Science,2014 (DOI: 10.1561/0400000042)
Dwork, C. & Roth, A., 'The Algorithmic Foundations of Differential Privacy', Foundations and Trends® in Theoretical Computer Science, 2014, DOI: 10.1561/0400000042
独立核验依据 · Independent verification: 核验依据 1 Crossref DOI 10.1561/0400000042
#差分隐私 #噪声