Networks Design Projects
COMPSCI 1440R
Subject & Catalog Number
Course Information
Description
This course examines how large language models (LLMs) can be applied to real-world design problems, including the generation of software and hardware (circuit) designs. Reasoning models now produce designs that are more often both correct and efficient. However, test-Time scaling raises inference cost, and generated designs must still be verified, for example with automatically generated testbenches.
Lectures cover reasoning LLMs, AI compute infrastructure, post-training (SFT, RLVR, GRPO), efficient inference (speculative decoding, quantization, KV-cache management), and agentic and multi-agent systems.
Students will:
- Complete assigned readings before each lecture.
- Study 12 research papers, present one in class, and review the other 11 (strengths, weaknesses, future directions).
- Take a midterm on the lectures and papers.
- Complete a team project (2–3 students) using LLM-assisted tools that contributes novel ideas, supported by an implementation and performance evaluation.
Course Notes
Preference will be given to upper-class undergraduates or graduate students majoring in Computer Science or Electrical Engineering. Please note that this course was previously listed as CS 144R.
Recommended Prep
Courses in Artificial Intelligence (AI), such as CS 1810 or CS 1820, as well as courses in AI computing, like CS 2420, and additionally, programming experience in languages such as Python or C++ and some research experience.
Available for Harvard Cross Registration
