What is Federated Learning Poisoning?

Federated learning architecture vulnerabilities: Byzantine attacks, model replacement, gradient manipulation, and techniques for poisoning global models through malicious participants.

What is Model Merging Exploits?

Exploiting model merging techniques (TIES, DARE, linear interpolation) and LoRA composition to introduce backdoors through individually benign model components.

What is Distributed Training?

Security vulnerabilities in multi-GPU, multi-node LLM training: gradient sharing attacks, parameter server compromise, insider threats, and infrastructure-level training exploits.

What is Emergence Exploitation?

How emergent capabilities create unpredictable security properties: testing for hidden capabilities, sleeper agent scenarios, deceptive alignment concerns, and capability elicitation.

What is Synthetic Data Attacks?

Attacks on synthetic data generation pipelines: model collapse from synthetic feedback loops, poisoning synthetic data generators, quality control bypass, and data provenance attacks.

What is Watermark Removal?

Techniques for removing AI watermarks: paraphrasing attacks, token substitution, embedding space perturbation, and implications for model provenance and accountability.

What is Lab: Federated Learning?

Hands-on lab implementing model poisoning attacks in a simulated federated learning setup using the Flower framework: Byzantine attacks, model replacement, and measuring attack impact.

What is Machine Unlearning Attacks?

Exploiting machine unlearning processes: recovering supposedly forgotten data, unlearning verification attacks, partial unlearning exploitation, and the fundamental limits of forgetting in neural networks.

What is Synthetic Data Risks?

Model collapse from training on synthetic data, quality degradation across generations, distribution narrowing, minority erasure, and strategies for safe synthetic data usage in LLM training.

What is Continual Learning Security?

Security risks in continual learning systems: catastrophic forgetting exploitation, task interference attacks, replay buffer poisoning, and stability-plasticity manipulation.

進階訓練漏洞

進階1 分鐘閱讀更新於 2026-03-15

AI 訓練中的進階安全威脅——涵蓋聯邦學習攻擊、模型合併風險、水印移除、合成資料投毒、遺忘攻擊與持續學習漏洞。

advanced federated-learning model-merging watermark synthetic-data unlearning

本節涵蓋超越標準資料投毒與微調攻擊的進階訓練漏洞。這些是針對 AI 訓練版圖中新興實務的攻擊：分散式訓練、模型合併、AI 產生訓練資料，以及知識移除。

進階訓練漏洞

進階1 分鐘閱讀更新於 2026-03-15

AI 訓練中的進階安全威脅——涵蓋聯邦學習攻擊、模型合併風險、水印移除、合成資料投毒、遺忘攻擊與持續學習漏洞。

advanced federated-learning model-merging watermark synthetic-data unlearning

進階訓練漏洞

聯邦學習攻擊

模型合併風險

水印移除

合成資料投毒

遺忘攻擊

持續學習漏洞

學習路徑

進階訓練漏洞

聯邦學習攻擊

模型合併風險

水印移除

合成資料投毒

遺忘攻擊

持續學習漏洞

學習路徑

進階訓練漏洞

學習路徑

相關文章

進階訓練漏洞

學習路徑

相關文章