Glossary#
- Accuracy#
Evaluation metrix measuring the fraction of predictions that a classification model got right (Wikipedia).
- Activation function#
A function (for example, ReLU or sigmoid) that generates a (typically nonlinear) output. Used in neural networks to define the output of a layer (Wikipedia).
- AUROC#
Area Under the ROC Curve. An aggregate measure of a classifier’ performance across all possible classification thresholds (Wikipedia).
- Automatic differentiation (autodiff)#
Soon!
- Backpropagation#
Mathematical technique used to obtain the gradients of the loss function with respect to the weights of a neural network. Applies the chain rule to compute the gradients one layer at a time, iterating backwards from the output layer (Wikipedia).
- Batch#
A set of examples used in one iteration (that is, one gradient update) during model training. Batch size defines the number of examples in a batch (Wikipedia).
- Bias#
Soon!
- Boosting#
Soon!
- Broadcasting#
Soon!
- Confusion matrix#
Soon!
- Decision boundary#
Frontier between the classes learned by a classification model.
- Precision#
Evaluation matrix measuring the frequency with which a model was correct when predicting the positive class (Wikipedia).
- Self-supervised learning#
Form of Machine Learning in which labels are automatically generated from the data itself, then a model is supervisely trained on the resulting dataset.