Trimming: Decoupling Multiplicative Depth from Modulus Chains in RNS-CKKS via Rational Levels
The episode discusses the paper 'Trimming,' which introduces a novel method for secure AI computation using homomorphic encryption. The authors detail how 'Rational Levels' allow researchers to independently optimize the structural depth and security parameters of complex models, making deep learning feasible for highly sensitive data like medical or financial records.