Loss functions

Loss functions are what is minimized by machine learning models during training.

Types

• Cross-entropy loss: for categorization or classification (e.g., sequence design)
• Focal loss: A modified version that emphasizes rare examples
• Frame aligned point error: A loss function for residues in 3D space
• Contrastive losses: Used to also manipulate the placement of embeddings with respect to each other

See also

• Fitness prediction using inverse folding can be improved with auxiliary B-factor and SASA losses
• Fine-tuning PLMs on rank is more effective than fine-tuning on property (regression)