Evaluating PQC KEMs, Combiners, and Cascade Encryption via Adaptive IND-CPA Testing Using Deep Learning

This project is a nascent academic reference implementation (9 days old, 0 stars) tied to a research paper. It attempts to apply deep learning to the challenge of validating ciphertext indistinguishability in PQC KEMs and hybrid 'combiner' constructions. While the methodology of using ML for cryptanalysis is an established niche, applying it as an automated 'adaptive IND-CPA' testing tool for PQC is a novel combination of disciplines. From a competitive standpoint, it lacks any moat; the value lies entirely in the research methodology rather than the codebase, which is easily reproducible by any security researcher with a standard ML stack. Frontier labs are unlikely to compete here as this is a specialized security auditing/verification tool. Its primary 'competitors' are traditional statistical test suites (NIST SP 800-22) and formal verification frameworks like Coq or F*, which offer mathematical proofs rather than the empirical, data-driven approximations this project provides. The low defensibility reflects the lack of community adoption and the commodity nature of the underlying binary classification models used for the testing.