Loss function calculates the loss for one row, while cost function does so for all target data points and aggregates them over N rows.Mean Squared Error (MSE) aggregates squared losses over N rows, providing a better evaluation metric.Accuracy in supervised machine learning (ML) measures correctly predicted instances in classification, not regression.Derivatives in uni-variable calculus show rate of change between two close points on a curve.Multi-variable calculus enables understanding dependencies among input variables in derivative calculations.Partial derivatives in multi-variable calculus help analyze model behavior with changing input variables.Chain rule in calculus is essential for finding derivatives of composite functions and is vital in backpropagation in ML.Training algorithm encompasses the entire model training process from initialization to parameter tuning.Optimization algorithm focuses on minimizing the loss function for model convergence, critical for training models.Model training includes parametric and non-parametric methods and various training algorithms.