Support multilabel binary cross entropy

[laggui]

The current implementation of BinaryCrossEntropyLoss only supports the binary case for 1-D inputs and targets. This PR extends the current definition to support Tensor<B, D>.

For multi-label classification the typical shape is [batch_size, num_classes] and targets are provided as multi-hot encoded targets.

Checklist

• Confirmed that run-checks all script has been executed.

Changes

Added support for multilabel BCE loss for input Tensor<B, D>

• Changed implementation to use log_sigmoid(x) instead of log(sigmoid(x)) with logits (numerical stability)
• Update label smoothing for multiclass (N > 2)
• Update loss class weighting to support multilabel case and Tensor<B, D>
• Fixed unit tests to use pre-computed values to compare against instead of comparing the function results with random values

⚠️ Breaking
Changed the default config to use BinaryCrossEntropyLoss without logits (expects probabilities, not raw predictions/logits). I think this makes more sense as a default and allows to use the config to build the loss with something more natural:

BinaryCrossEntropyLossConfig::new().with_logits(true)

This is a bit of an opiniated change, could be reverted if really desired.

Testing

Added unit tests for new multi-label case and updated the binary 1-D tests.

[codecov]

Codecov Report

Attention: Patch coverage is 99.50739% with 1 lines in your changes are missing coverage. Please review.

Project coverage is 86.53%. Comparing base (b0c5986) to head (b498d6c).

Files	Patch %	Lines
...ates/burn-core/src/nn/loss/binary_cross_entropy.rs	99.50%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1571      +/-   ##
==========================================
+ Coverage   86.48%   86.53%   +0.05%     
==========================================
  Files         684      684              
  Lines       78093    78248     +155     
==========================================
+ Hits        67535    67713     +178     
+ Misses      10558    10535      -23     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[louisfd]

LGTM

[antimora]

LGTM