• Overfit a single batch
• Run with a high number of epochs
• Set seeds
• Rebalance the dataset
• Use a neutral class
• Set the bias of the output layer
• Deliberately overfit in case of physical simulations
• Tune the learning rate
• Use fast data pipelines
• Use data augmentation
• Train an AutoEncoder on unlabeled data, use latent space representation as embedding
• Utilize embeddings from other models
• Use embeddings to shrink data
• Use checkpointing
• Write custom training loops
• Set hyperparameters appropriately
• Use EarlyStopping
• Use transfer-learning
• Employ data-parallel training
• Use sigmoid activation for multi-label tasks