Labs · Training

Cross-entropy against MSE

The same batch of predictions scored two ways — how hard each loss punishes a confident mistake, what a batch mean hides, and which positions a mask lets count.

L=−1∣S∣∑i∈Slog⁡pi[yi]L = -\frac{1}{|S|}\sum_{i \in S} \log p_i[y_i]5 classes · a batch of 4

Four examples: confidently right, confidently wrong, unsure, near-tie.

What the two losses say about this batch

Reads /api/labs/loss — loading

Computing…

Cross-entropy against MSE — Labs — TransformerLab