Trustworthy AI for SWE

Basic Info

Instructor: Dr. Bowen Xu (bxu22@ncsu.edu)

TA: Weiyuan Ding (wding8@ncsu.edu)

TA: Hua Yang (hyang45@ncsu.edu)

Semester: Fall 2026

Schedule: Mon/Wed, 3:00 PM - 4:15 PM

Location: 01226 Koch Hall (formerly Engineering Building 2)

Office Hours: Tue, 2:00 PM - 3:00 PM, Koch Hall (formerly Engineering Building II (EB2)), Room 3228

Resources

Syllabus Assignments Gradescope Discord

Course Outline (Tentative)

Week	Date	Topic	Materials	Notes
Class 1	Mon Aug 17	Course Intro	[Slides]	—
Class 2	Wed Aug 19	Data-centric TAI4SE	[Slides]	[The Devil is in the Tails: How Long-Tailed Code Distributions Impact Large Language Models]
Class 3 (No class due to school event)	Mon Aug 24	Replaced with Data Quality Issues in Code Corpora Assignment (See Assignment section)	—	—
Class 4	Wed Aug 26	Group project demos	-	—
Deadline	Sun 30 Aug	Team Formation	NA	Submit Team Formation form (one representative member per team should submit the form on behalf of the team)
Class 5	Mon Aug 31	AI Coding Workflow Challenge (In-class activity)	[Slides]	—
Class 6	Wed Sep 2	Code Representation	[Slides]	[Stanford CME295 Transformers & LLMs ]
Class 7	Wed Sep 9	Guest Lecture	[Slides]	Guest speaker: Andre Lustosa (Red Hat)
Class 8	Mon Sep 14	Guest Lecture	—	Guest speaker: Erkang Zheng (Founder, Ariso + JupiterOne)
Class 9	Wed Sep 16	Group Project Consultant	—	No class — group project consultant sessions
Class 10	Mon Sep 21	Guest Lecture	[Slides]	Guest speaker: Rameez Malik Talk: The Advent of Neurosymbolic AI within Regulatory Affairs
Class 11	Wed Sep 23	Group Project Consultant	—	No class — group project consultant sessions
Class 12	Mon Sep 28	Assignment Discussion	—	—
Class 13	Wed Sep 30	Robustness of AI in Coding	[Slides]	—
Class 14	Mon Oct 5	Guest Lecture	[Lean Intelligence: GitHub Copilot Data Science Efforts]	Guest speaker: Rahul Pandita (GitHub) Talk: Lean Intelligence: GitHub Copilot Data Science Efforts Extra reading: [Quantifying GitHub Copilot's impact on developer productivity and happiness]
Class 15	Wed Oct 7	Model Quantization	[Slides]	Saima Afrin (William & Mary)
Class 16	Mon Oct 12	Group Project Proposal Presentations	[Proposal Slide Template]	In-class group project proposal presentations
Class 17	Wed Oct 14	Group Project Proposal Presentations	[Proposal Slide Template]	In-class group project proposal presentations
Class 18	Wed Oct 21	—	—	—
Class 19	Mon Oct 26	Code Backdoor Attack and Defense	[Slides]	[A small number of samples can poison LLMs of any size]
Class 20	Wed Oct 28	Guest Lecture	—	Guest speaker: Toufique Ahmed (IBM) Talk: Reproduction Test Generation for SWE Issues: Methods and Challenges
Class 21	Mon Nov 2	Basics of Causal Techniques	Causal Inference Module These four classes teach how to use causal inference to interpret the security of a model. Every session includes in-class activities and quizzes (they count toward Class activities). Competition Rounds 1–3 run alongside. Details →
Class 22	Wed Nov 4	Causal Analysis for Interpreting AI Behaviour
Class 23	Mon Nov 9	Lab 1: Causal Effect Estimation
Class 24	Wed Nov 11	Lab 2: Causal Discovery
Class 25	Mon Nov 16	—	—	—
Class 26	Wed Nov 18	—	—	—
Class 27	Mon Nov 23	Final Group Project Presentations	[Final Presentation Template]	In-class final group project presentations
Class 28	Mon Nov 30	Final Group Project Presentations	[Final Presentation Template]	In-class final group project presentations

Assignments (Tentative)

Assignment	Due	Materials	Notes
Data Quality Issues in Code Corpora	Mon Sep 7, 2026, 3:00 PM	[Assignment]	[Notes]
Code Representation Learning	Wed Sep 16, 2026, 3:00 PM	[Assignment]	[Notes]
Does Quantization Make Code LLMs Brittle?	Wed Oct 28, 2026, 3:00 PM	[Assignment]	—
EVALOOP: Robustness of LLMs in Code Generation	Wed Nov 4, 2026, 3:00 PM	[Assignment]	—

Assignments will be submitted through Gradescope. Detailed instructions will be posted here as they are released.

Causal Inference & Code Backdoor Detection Module

This module

Getting started & Round 1 Wed Oct 28, 9:00 AM – Mon Nov 2, 3:00 PM

Consent questionnaire Background Survey Competition Round 1 unlocks after the Background Survey Class 1, Mon Nov 2

Causal classes & Round 2 Mon Nov 2, 3:00 PM – Wed Nov 11, 5:00 PM

In-class quizzes Round 2 (Nov 4, 5 PM – Nov 9, 3 PM) Round 2 submission form

Round 3 & final report Wed Nov 11, 5:00 PM – Wed Nov 18, 3:00 PM

In-class quiz Round 3 Round 3 submission form Final report

From late October to mid-November, the course runs a module that pairs four classes on causal inference with a private Kaggle competition on detecting backdoors (poisoned training samples) in code models. The classes introduce causal inference: telling whether one thing actually causes another, rather than just appearing alongside it. That skill is widely useful in software engineering, machine learning, and data analysis.

[Module overview slides]: what you will do in this module and how it is graded.

Timeline

All dates are in 2026 and all times are Eastern Time (ET).

Wed Oct 28, 9:00 AM

Module starts: study code, Background Survey, and Competition Round 1

You receive your study code and a short consent questionnaire by email from TA Hua Yang. Complete the Background Survey (about 10 minutes) first: the Kaggle invitation link is at the end of it. Round 1 closes Mon Nov 2, 3:00 PM. Each round ends with a short submission form (survey).
Mon Nov 2

Class 1 on causal inference

Round 1 closes at 3:00 PM.
Wed Nov 4

Class 2; Competition Round 2 opens

Round 2 opens at 5:00 PM and closes Mon Nov 9, 3:00 PM.
Mon Nov 9

Class 3

Round 2 closes at 3:00 PM.
Wed Nov 11

Class 4; Competition Round 3 (final) opens

Round 3 opens at 5:00 PM and closes Wed Nov 18, 3:00 PM.
Wed Nov 18, 3:00 PM

Round 3 closes; module ends

Submit the Round 3 submission form.

What you do

Background Survey

About 10 minutes. The Kaggle invitation link is at the end of the survey.

[Open the Background Survey]

Kaggle competition

Detect poisoned training samples (backdoors) in code models, in three rounds:

Round 1: Wed Oct 28, 9:00 AM – Mon Nov 2, 3:00 PM
Round 2: Wed Nov 4, 5:00 PM – Mon Nov 9, 3:00 PM
Round 3: Wed Nov 11, 5:00 PM – Wed Nov 18, 3:00 PM

Complete the Background Survey to get the invitation link. Each round ends with a short submission form. Set your Kaggle team name to your study code.

In-class quizzes

Quizzes during the four causal inference classes. They can only be taken in class.

Final report

A final report at the end of the module. Due date to be announced.

Use your study code, not your name or Unity ID, on the surveys, on the quizzes, and as your Kaggle team name.

Grading

Class activities or seminar participation (20%): this module counts here, 20 points in total:
- In-class quizzes: 4 × 2 points = 8 points
- Competition rounds: 3 × 3 points = 9 points
- Final report: 3 points
Your grade for each competition round reflects both your leaderboard results and your participation throughout the round (valid submissions during the round and its submission form). Quizzes are graded on both your answers and your participation.

Research study

This module is also part of a research study. Every student does all activities as part of the course; each student chooses whether their data may be used. If you join, you earn 5 points of extra credit by completing the study (staying in it through the end of the module), and you receive a summary of the findings. If you do not join, or withdraw before the end, you can earn the 5 points by recording a 3-minute video introducing one backdoor detection method. The instructor does not know who joined until final grades are posted. Details will come by email.

Trustworthy AI for Software Engineering