OA4128 Large Language Models for Defense Applications

This course introduces students to the use of Large Language Models (LLMs) for analysis of military problems. This course begins with a treatment of the underlying mathematics and theory of LLMs, giving an overview of the fundamental LLM building blocks and working towards an understanding the foundational principles and architectures of LLMs and how they produce their inferences. The course builds on the conceptual foundations by focusing on practical applications of how LLM frameworks can provide an objective, trusted, and relevant basis for military leaders to make informed decisions. The course focuses on modern transformer-based LLM architectures, testing and evaluation methods, and framework enhancements such as retrieval augmented generation (RAG) methods, agentic workflows, and methods for model fine-tuning and evaluation. Throughout the course other topics are also covered including LLM deployment considerations, model quantization and sharding, and other alignment, safety, security, and bias considerations. The course concludes with students addressing a real-world military analytic problem that leverages LLMs where students select and build an open source LLM pipeline, on their computers or a remote server, and experiment with various prompts, parameters, and RAG and/or agentic frameworks.

Cross Listed Courses

To be cross listed as CS4325

Prerequisite

OA3801: Computation Methods for Operations Research II or equivalent

Lecture Hours

3

Lab Hours

2

Course Learning Outcomes

Upon successful completion of the course the student will:

  • Understand the foundational principles and architectures of LLMs.

  • Explore a range of LLM applications across various domains.

  • Gain practical skills in using frameworks and tools for optimizing LLM inferences.

  • Learn RAG techniques for combining retrieval systems with LLMs.

  • Understand agentic frameworks and their applications in autonomous AI.

  • Develop capabilities in prompt engineering, RAG, and agentic frameworks for specific use cases.

  • Assess risks, safety, security, and ethics concerns for LLM use in defense analysis applications.