Loss Functions

32 problems

• MSE loss + backwardEasy
• CLIP InfoNCE LossMedium
• Contrastive lossMedium
• Dice lossEasy
• BCE from logits — the stable sigmoid + BCE pairingMedium
• BCE — clip the probabilities so log stays finiteEasy
• Loss — pick the better modelEasy
• DPO lossMedium
• Focal loss (multiclass)Medium
• Huber & Hinge lossesEasy
• IPO / SimPO LossMedium
• KL divergence (discrete)Easy
• Knowledge-Distillation LossMedium
• KTO (Kahneman-Tversky) LossHard
• Label-smoothed cross-entropyMedium
• Linear regression — sum of squared residuals (SSR)Easy
• ListNet Listwise Ranking LossMedium
• Log-softmaxEasy
• Logistic regression — the BCE gradient (error × feature)Easy
• Logistic regression — binary cross-entropy lossEasy
• Logistic regression — likelihood of the whole datasetEasy
• Multi-class cross-entropy lossEasy
• ORPO Loss (Odds-Ratio Preference Optimisation)Medium
• Pairwise Ranking Loss (BPR / RankNet)Medium
• Pinball (Quantile) Regression LossMedium
• Reward-Model Margin LossMedium
• Ridge regression lossEasy
• Loss over time — sum the per-step lossesEasy
• In-Batch Sampled Softmax + logQ CorrectionHard
• Triplet lossMedium
• VAE ELBO lossMedium
• Weighted cross-entropyEasy