LATIDIA · Ciberseguridad

Rompiendo el esquema $\mathsf{TraintoDefend}$ mediante la agrupación adaptativa de neuronas en la extracción de modelos criptoanalíticos.

arXiv: 2611.07370v1Tipo de anuncio: nuevo Resumen: La extracción del modelo criptoanalítico representa una amenaza fundamental para la confidencialidad del modelo de aprendizaje profundo al permitir la recuperación precisa de parámetros con garantías teóricas de funct

La noticia

arXiv:2611.07370v1 Announce Type: new Abstract: Cryptanalytic model extraction poses a fundamental threat to deep learning model confidentiality by enabling precise parameter recovery with theoretical guarantees of functional equivalence. To counter this growing threat, $\mathsf{TraintoDefend}$ (NeurIPS 2025) was proposed as the first dedicated defense against cryptanalytic extraction and remains the only defense believed to be effective to date. It encourages neurons to learn similar weights so that they switch activation states together, making their individual contributions harder to distinguish. However, we notice that neurons with similar weights need not switch activation states together under small perturbations. Instead, weight similarity can cause the original clustering rule to merge signatures from distinct neurons