Module loss 

Differentiable loss functions for neural network training.

These loss functions work with autograd Tensors and support backpropagation for gradient-based optimization.

Example

use aprender::nn::loss::{MSELoss, CrossEntropyLoss};
use aprender::autograd::Tensor;

// Regression loss
let criterion = MSELoss::new();
let pred = Tensor::from_slice(&[1.0, 2.0, 3.0]).requires_grad();
let target = Tensor::from_slice(&[1.1, 2.0, 2.9]);
let loss = criterion.forward(&pred, &target);
loss.backward();

// Classification loss
let criterion = CrossEntropyLoss::new();
let logits = Tensor::new(&[1.0, 2.0, 0.5, 0.1, 3.0, 0.2], &[2, 3]).requires_grad();
let targets = Tensor::from_slice(&[1.0, 2.0]);  // class indices
let loss = criterion.forward(&logits, &targets);

References

• Bishop, C. M. (2006). Pattern Recognition and Machine Learning. Springer.

Structs

BCEWithLogitsLoss

Binary Cross-Entropy with Logits loss.

CrossEntropyLoss

Cross-Entropy Loss for classification.

L1Loss

Mean Absolute Error loss for regression.

MSELoss

Mean Squared Error loss for regression.

NLLLoss

Negative Log Likelihood loss.

SmoothL1Loss

Smooth L1 Loss (Huber Loss).

Enums

Reduction

Reduction mode for loss functions.