Model Building and Training

This unit guides students through constructing, training, evaluating, and tuning TensorFlow models for classification tasks. It covers tensor transformations, input pipelines, estimators, training mechanics (backpropagation, sessions, variables), visualization with TensorBoard, and using Google Cloud tools for scalable experiments and storage.

Learning Objectives

• Apply TensorFlow tensor transformations, image operations, and input functions to preprocess and prepare datasets for model training
• Demonstrate building, training, and evaluating binary and multinomial logistic regression models and deep neural networks using estimators, variables, dropout layers, and backpropagation
• Analyze training dynamics and model performance using TensorBoard, session inspection, and hyperparameter tuning to detect and mitigate overfitting
• Evaluate reproducibility and scalability by running Estimator training jobs and managing artifacts with gcloud and Google Cloud Storage