Machine learning for modular multiplication
This work addresses the challenge of applying ML to cryptographic tasks, but it is incremental as it confirms existing hardness assumptions without breakthroughs.
The paper tackled the problem of learning modular multiplication using machine learning, specifically testing circular regression and a transformer model, but found limited success, providing evidence for the hardness of such tasks in cryptography.
Motivated by cryptographic applications, we investigate two machine learning approaches to modular multiplication: namely circular regression and a sequence-to-sequence transformer model. The limited success of both methods demonstrated in our results gives evidence for the hardness of tasks involving modular multiplication upon which cryptosystems are based.