Machine Learning Model Walkthroughs

This Machine Learning Model Walkthrough Activity will help you:

• See the required components for two common ML image processing models: logistic regression model and ResNet50 transfer learning model. 

• Run two common ML image processing models on the same dataset (cat and dog classification) and understand the difference in performance between them. 

• Understand each of the components present within the logistic regression model and ResNet50 transfer learning model. 

Required Materials/Resources:

• Access to a computer and large screen (if you want share with others)

• A Google account to access Google Colab

Keywords:

Logistic Regression Model - A statistical model used in predicting the probability of a certain outcome based on a determined relationship between one or more variables, or predictors. In other words, the logistic regression model provides a categorical outcome based on a combination of input variables. 

Transfer Learning Model - A machine learning model that uses a model which has been previously trained for a separate task to be used on a new, related task. In contrast to creating a completely new model, a transfer learning model has imported weights from its previous training and adapts these weights to better fit the new task at hand. Using a transfer learning model requires less training data and computational power.

ResNet50 Model - A well-known deep learning model composed of 50 layers that utilizes residual learning techniques to train its deep network efficiently. ResNet was created in 2015 by the Microsoft Research team. The ResNet50 model has been trained on the ImageNet image dataset for classification of more than 1,000 different objects. 

ImageNet - An image dataset initially created by a group of researchers at Stanford and Princeton which has grown to contain over 14 million images representing about 80,000 different meaningful concepts, referred to as “synonym sets”.