Neural Engine Byte NEB

Neural Engine Byte (NEB) is an open-source Layer 1 cryptocurrency project exploring a different approach to Proof-of-Work mining. Instead of relying exclusively on conventional hash computations, NEB integrates deterministic neural-network training directly into its mining process.

The protocol uses a compact 784-128-10 neural network with 4-bit weights, trained on samples from the MNIST handwritten-digit dataset. Each mining attempt selects 16 samples using chain-defined rules and performs an integer-based training step. The resulting gradients contribute to a BLAKE3-derived mining ticket, allowing network nodes to independently reproduce and verify the computation.

NEB introduces a share-based training mechanism in which qualifying mining attempts contribute gradients toward updating the network's shared model. These contributions are aggregated and applied with a one-block delay, allowing the model to evolve according to the blockchain's recorded activity.

The blockchain uses a UTXO transaction model with Ed25519 signatures and BLAKE3 hashing. Its published testnet parameters specify a 60-second target block interval, an initial reward of 20 NEB, and halvings every 450,000 blocks. Five percent of each block reward is allocated to the project, with the remainder going to miners.

The initial public testnet is documented under software version 0.1.0, with support for CPU mining, NVIDIA GPU mining, node operation and command-line wallet management. The implementation is publicly available under the Apache-2.0 license.

Neural Engine Byte remains in the testnet stage, and its testnet coins have no stated monetary value. The project aims to demonstrate how verifiable neural-network training can be integrated into blockchain consensus without relying on trusted external computation.