LATIDIA · Ciberseguridad

SANAI: Paridad con clave distribuida ECC segura integrada en los pesos para proteger las redes neuronales contra ataques de inversión de bits.

arXiv:2611.02042v1 Anuncio Tipo: nuevo Resumen: Las redes neuronales cuantizadas almacenan pesos en memoria densa, donde los cambios de bits causados ​​por el escalado de voltaje, el envejecimiento, la radiación o ataques deliberados como RowHammer pueden afectar gravemente

La noticia

arXiv:2611.02042v1 Announce Type: new Abstract: Quantized neural networks store weights in dense memory, where bit flips caused by voltage scaling, aging, radiation, or deliberate attacks such as RowHammer can severely degrade model behavior. A single corrupted high-order bit can cause misclassifications or nonsensical language-model outputs. Embedded error-correcting codes (ECC) avoid dedicated parity storage by hiding parity within model weights, but existing schemes embed uniformly across layers and bit planes, leave parity unsecured, and tolerate bit-error rates (BERs) only up to about 6%. An adversary aware of sensitive weights and the ECC scheme can also bypass protection by exceeding the correction capability of targeted codewords. We propose SANAI, a sensitivity-guided, bit-plane-selective