You are on page 1of 17

Selected Topics in AI

By. Dr. Eman Ahmed


e.ahmed@fci-cu.edu.eg
Group?
Course Contents
• Revision on Logistic Regression and Neural Networks.
• Design Issues (Initialization, activation function choice, regularization,
loss functions, …)
• Deep Learning Models (CNNs) and its types.
• Transfer Learning and pretrained models.
• Performance Measures (Confusion matrix)
• Discriminative vs Generative Learning.
• Project (Applying deep learning and transfer learning in different
applications) – Optimization?
Linear Regression

Binary Classification
• Classify an image into 2 classes.
• Class Cat : 1
• Class Dog : 0
Logistic Regression

Sigmoid Function
• For large values of z:
• Sigmoid approaches 1
• For very small values of z:
• Sigmoid approaches zero
• When z=0:
• Sigmoid outputs 0.5
Loss Function

Convex Function

We would like the loss function to be convex to have one minimum.


This is not the case for non-convex functions, they have many local minima.
Loss Function of Logistic Regression

Cost Funtion

Gradient Descent
W increases

W decreases
Logistic Regression
Delta learning rules

Thanks

You might also like