[archived] Lecture 1.4: From logistic regression to fully-connected networks

In the final video of this introductory lecture, we start with the logistic regression model. We outline the learning objective (i.e., the log-likelihood function), and the gradient-based learning algorithm. Further, we indicate that the logistic regression could be seen as a neuron, and stacking multiple logistic regressors result in a hierarchical model. We refer to that model as fully-connected neural network.

[donnap]

He is right. It is clearer to look at the slides than to listen to his explanation, unfortunately.