I got a bit confused about the usage of machine learning terminology in books / papers / discussions that seems somehow not completely consistent to me. Therefore, I want to know if you would agree with my understanding of the following terms (if not I would appreciate any suggestion for improvement). I come from a NLP background, so I hope it's ok to use this as example.

We have machine learning methods (e.g. BERT) that define a way to solve a problem / multiple problems by using a trained model. The model has an architecture (e.g LSTM) that defines its "building blocks" and a way how it is trained to be useful to solve the problem. This way is either referred to as a task that the model has to perform or as an objective that is used to train the model (e.g. language model objective). The model typically learns from this task or objective by using a loss function and backpropagation to adapt its values.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.