DPNE-VVTPM (Dynamic Parallelized Noise-Enhanced Vector Valued Tree Parity Machine) for Secure Cybersecurity
Securing key exchange over public channels is a foundational requirement of modern cybersecurity, yet classical schemes such as RSA, AES and DSA face growing computational and adversarial pressure. Neural cryptography with Tree Parity Machines (TPM) establishes shared keys through mutual learning, but existing Vector Valued TPM (VVTPM) designs suffer from slow synchronization, static parameterization and vulnerability to synchronization following attacks. This paper proposes DPNE-VVTPM, a Dynamic Parallelized Noise Enhanced VVTPM framework that integrates four cooperating mechanisms: an error driven dynamic weight range schedule, secret Gaussian noise injection during synchronization, parallel synchronization of multiple TPM lanes, and SHA-256 based high entropy key derivation. Simulation studies show that dynamic range scheduling reduces mean synchronization effort from 2128 to 188 iterations, an 11.3 times acceleration, while noise injection lowers the weight agreement achieved by a geometric eavesdropper from 33.2 percent to 19.3 percent. The derived 256 bit session keys attain 0.997 bits per bit Shannon entropy, 99.0 percent monobit and 99.7 percent runs test pass rates, and a 50.06 percent avalanche effect. End to end benchmarking against RSA, AES and DSA records the highest throughput of 3237708.35 bits per second, the lowest encryption time of 2.72e-05 seconds and decryption time of 5.48e-06 seconds, with marginally lower memory usage, confirming DPNE-VVTPM as an efficient and resilient key exchange primitive.