What are the commonly used loss functions? (Multiple choice)

Question

Options

• AMean variance
• BSigmoid cross entropy
• CSoft max cross enl10py
• DSparse cross entropy
• EWeighted S1gmo1d Cross Entropy

• A
  100% (52)

Explanation

All five options represent valid, commonly used loss functions in machine learning - this is a "select all that apply" question with no distractors.

• A (Mean Variance) captures the spread of predictions and is used in uncertainty-aware and risk-sensitive models.
• B (Sigmoid Cross Entropy) is the standard loss for binary classification, pairing a sigmoid activation with cross entropy to output probabilities between 0 and 1.
• C (Softmax Cross Entropy) extends this to multi-class classification, normalizing logits across all classes before computing the log loss.
• D (Sparse Cross Entropy) is a memory-efficient variant of softmax cross entropy that accepts integer class labels directly instead of one-hot encoded vectors.
• E (Weighted Sigmoid Cross Entropy) adjusts B by applying per-class weights, making it essential for imbalanced datasets where one class dominates.

Memory tip: Think of them as a family tree - Sigmoid CE (binary) → Weighted Sigmoid CE (imbalanced binary) → Softmax CE (multi-class) → Sparse CE (multi-class with integer labels). Mean Variance stands alone as the odd one out focused on prediction spread, not class labels. Remembering the binary-to-multiclass progression covers four of the five answers instantly.

No community discussion yet for this question.