Masahiro Kaneko · 2026 · arXiv
DOI: 10.48550/arXiv.2610.28045 · open-access PDF
신경 네트워크에 대한 백도어 공격으로부터 방어하기 위해, 방어자는 입력이 트리거인지 여부를 결정하는 검증기 기능을 식별해야 하며, 단독으로 개별 트리거를 거부하는 것이 아니라, 공격자가 동일한 검증기를 차단할 수 있는 것보다 더 빨리 만족시키는 새로운 트리거를 생성할 수 있기 때문이다.If a backdoor verifier could be installed with the asymmetry of a cryptographic authentication scheme, in which only the attacker can produce trigger input while the defender cannot reverse engineer the verifier from observations, learning-based defense would be impossible in principle.
Machine translation (Meta M2M100); the English original follows.
To defend against backdoor attacks on neural networks, the defender must identify the verifier function that determines whether an input is a trigger, rather than rejecting individual triggers in isolation, since an attacker can generate new triggers satisfying the same verifier faster than they can be blocked. If a backdoor verifier could be installed with the asymmetry of a cryptographic authentication scheme, in which only the attacker can produce trigger inputs while the defender cannot reverse-engineer the verifier from observations, learning-based defense would be impossible in principle
Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese
GET https://store.agentexchange.work/gw/papers/verify?title=...&year=... (0.003 USD), fetch this record as JSON with /gw/papers/paper?id=2610.28045, or list what it cites with /gw/papers/references?id=.... Pay per call in USDC (x402) or with a 19 USD card key for 1,000 calls. Catalogue · llms.txtSources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange