Module gradient_noise 

Gradient noise injection for training regularization.

This module provides configurable noise injection into gradient tensors, useful as a regularization technique during neural network training. Supported noise types include Gaussian, Uniform, Laplacian, and Scheduled Gaussian (where noise magnitude decays over training steps).

Example

use ipfrs_tensorlogic::gradient_noise::{
    GradientNoiseConfig, GradientNoiseInjector, NoiseType,
};

let config = GradientNoiseConfig {
    noise_type: NoiseType::Gaussian,
    initial_scale: 0.01,
    decay_rate: 0.0,
    clip_value: Some(0.05),
    seed: 42,
};

let mut injector = GradientNoiseInjector::new(config);
let mut gradients = vec![1.0, 2.0, 3.0, 4.0];
injector.inject(&mut gradients);
// gradients now contain added noise

Structs

GradientNoiseConfig

Configuration for gradient noise injection.

GradientNoiseInjector

Injects configurable noise into gradient arrays for training regularization.

NoiseSample

A batch of noise samples together with summary statistics.

NoiseStats

Aggregate statistics about noise injections performed by an injector.

Enums

NoiseType

The type of noise distribution to inject into gradients.