This paper presents NeuroMAC — a message authentication code based on a neural network architecture (8.6 KB of weights) with innovative batch processing and caching mechanisms. The MAC generator delivers a performance of 0.41 μs/MAC in batch mode (174× acceleration) while maintaining full cryptographic strength: 15/15 NIST SP 800-22 tests passed on 838 sequences, quantum resistance (256 bits against Grover's algorithm), avalanche effect of 49.94%, ideal 0/1 balance (50.00% ±0.05%), and no collisions (0/100000). Cryptanalysis has been performed, including differential and linear cryptanalysis, timing attacks, and forgery attacks. All tests confirmed the cryptographic strength of the algorithm.
Maria Surkova (Sun,) studied this question.