Deep Learning Model Engineering

This unit focuses on engineering practical deep learning systems: implementing models in frameworks (TensorFlow, Chainer, Torch), optimizing learning and inference, and hardening models for real-world use across vision, language, and behavioral applications. It builds on foundational theory from earlier units and prepares students to create, evaluate, and deploy robust, high-performance deep learning applications in the culminating unit.

Learning Objectives

• Analyze and compare optimization and probability techniques (e.g., gradient methods, loss formulations, probabilistic outputs) to select appropriate training strategies for given model tasks
• Apply deep learning frameworks (TensorFlow, Chainer, Torch) to implement networks for image, text, and real-time behavior tasks and measure processing and inference performance
• Demonstrate methods to mitigate generalization and security issues by implementing regularization, feature selection, adversarial defenses, and online/one-shot learning approaches
• Evaluate trade-offs involved in creating deployable applications by testing model accuracy, latency, resource use, and user-support considerations, and propose improvements for production readiness