Building and Training Neural Models

This unit guides learners through transforming and visualizing data, constructing TensorFlow models, and training both logistic and deep neural networks. Students will implement operations, variables, input functions and estimators, use TensorBoard and cloud utilities, and apply regularization and tuning techniques to improve model performance.

Learning Objectives

• Analyze tensor transformations and image preprocessing techniques to prepare datasets for binary and multinomial classification tasks
• Apply TensorFlow primitives (Operations, Variables, Sessions, Input Functions) and the Estimator API to build, run, and deploy models locally and to Google Cloud Storage using gcloud
• Demonstrate training workflows including backpropagation, adding dropout and adjusting network depth, and visualize training and debugging information with TensorBoard
• Evaluate model performance for binary and multiclass classifiers, iterate with hyperparameter tuning and dataset augmentation, and document improvements using reproducible training pipelines