Struct NeuralPathValidator 

pub struct NeuralPathValidator {
    pub layer_widths: Vec<usize>,
}

Validates the neural inference path of a tensor computation.

Models the tensor operation as a small feedforward network. The “path” is the binary activation pattern at each layer (ReLU > 0 = 1, else 0). The miner must find a nonce that makes the actual path match an expected path signature (derived from the challenge) within a Hamming distance.

Fields

layer_widths: Vec<usize>

Layer widths for the feedforward network

Implementations

impl NeuralPathValidator

pub fn expected_path_signature(&self, challenge_hash: &str) -> Vec<u8>

Derive the expected path signature from the challenge hash. Returns a bit vector (as Vec of 0/1 values) representing the expected activations.

pub fn compute_actual_path( &self, tensor: &Tensor, nonce: u64, ) -> Result<Vec<u8>, TribeError>

Compute the actual activation path for a tensor with a given nonce. Simulates a feedforward pass: input -> (linear + ReLU) per layer. Returns the binary activation pattern.

pub fn hamming_distance(a: &[u8], b: &[u8]) -> u32

Compute Hamming distance between two bit vectors.

pub fn validate( &self, actual_path: &[u8], challenge_hash: &str, max_distance: u32, ) -> bool

Validate that the actual path is close enough to the expected path.

pub fn path_to_hex(path: &[u8]) -> String

Encode path bits as a compact hex string for on-chain storage.

Trait Implementations

impl Default for NeuralPathValidator

fn default() -> NeuralPathValidator

Returns the “default value” for a type.