Repurposing Neural Networks for Efficient Cryptographic Computation
Xin Jin
Network and Distributed System Security (NDSS) Symposium 2025 · Day 1 · Privacy & Cryptography 1
In an era increasingly reliant on robust digital security, the performance of cryptographic operations remains a critical bottleneck. This talk, "Repurposing Neural Networks for Efficient Cryptographic Computation," presented by Xin Jin at the NDSS Symposium, introduces **TensorCrypt**, a novel framework that leverages the inherent computational graph properties of neural networks to significantly accelerate cryptographic computations. Instead of traditional machine learning approaches that attempt to *train* a model for cryptographic functions, TensorCrypt directly transforms cryptographic algorithms into neural network computational graphs, effectively repurposing existing AI hardware and software stacks for high-speed encryption and decryption.
AI review
TensorCrypt is a genuinely novel systems paper that earns its place at NDSS: the core insight — treat crypto algorithms as computational graphs and compile them onto NN accelerator stacks without any training — is non-obvious and technically substantive. 5.4x over optimized GPU baselines for TLS 1.3 ciphers is a real number, and the formal security equivalence proof shows the authors understood they were playing with fire near the crypto/ML boundary. Not a 5-star because the implementation side-channel story is left largely open and the deployment diversity claim leans more on the NN…