Research

My research interests are at the intersection of Software Engineering and Artificial Intelligence (AI). I study how to build dependable learning-enabled software systems when AI models are integrated into real software pipelines. Rather than treating models as opaque libraries, I view them as probabilistic components whose use must be dynamically governed by software logic. This requires software to decide when to rely on models versus deterministic code, and how to validate and safely integrate model outputs, including model-generated code artifacts.

I organize this agenda across three layers of the AI-enabled software stack. The model layer focuses on the behavior of model outputs. The runtime layer focuses on deployment efficiency when model execution is adaptive. The infrastructure layer focuses on the correctness of the GPU and tile-program execution ecosystem that runs these workloads.

Behavioral reliability at the model layer: uncertainty-aware selective prediction and input adaptation that let systems decide when to accept an output, when to abstain, and how to adapt what the model sees under controlled steps.
Efficiency robustness at the runtime layer: analysis methods that characterize cost variability from adaptive model behavior and techniques that target predictable worst-case deployment cost.
Infrastructure reliability at the infrastructure layer: bug taxonomies and automated testing tools for GPU and tile-program compilation and execution, to detect hidden bugs.

Behavioral reliability at the model layer: uncertainty-aware selective prediction and input adaptation that let systems decide when to accept an output, when to abstain, and how to adapt what the model sees under controlled steps. Specially tailored to imrpove Code-LLMs.

Efficiency robustness at the runtime layer: analysis methods that characterize cost variability from adaptive model behavior and techniques that target predictable worst-case deployment cost.

Infrastructure reliability at the infrastructure layer: bug taxonomies and automated testing tools for GPU and tile-program compilation and execution, to detect hidden bugs.

Research Advising and Mentoring

Research Advising

I have advised two undergraduate students on research published at top-tier software engineering conferences (ICSE, ISSTA).

Tingxi Li (PhD, UT Dallas, Fall 2024-now): Published USENIX SEC 2025
Zexin Xu (PhD, UT Dallas, Fall 2024-now): Published USENIX SEC 2025
Nidhi Majoju (BS, UT Dallas, Summer 2024-now) : Published ISSTA 2025
Aaryaa Moharir (BS, UT Dallas, Spring 2025-now)
Laura Tartar (BS, UT Dallas, Summer 2024-Spring 2025)
Ishan Patel (MS, UT Dallas, Fall 2024-Spring 2025)
Zijie Zhao (BS, Tianjin University, Summer 2023- Summer 2024): Published ICSE 2025

Research Mentoring

Research, Inquiry, Design Experience (RIDE) Projects UT Dallas, 17 students (four freshman, eight sophomore, five junior) Spring 2025
Research, Inquiry, Design Experience (RIDE) Projects UT Dallas - 12 students (three freshman, two sophomore, five junior, two senior), Fall 2024
Summer High School Interns hosted by UT Dallas - three students in Summer 2024, two students in Summer 2025
Clark Summer Research Program hosted by UT Dallas - one student in Summer 2023, three students in Summer 2024, three students in Summer 2025
Undergraduate Research Apprenticeship Program (URAP) hosted by UT Dallas- two students in Fall 2024, two students in Spring 2025
Summer Platform for Undergraduate Research (SPUR) - four students in Summer 2024

Presentations

Conference Talks and Posters

SoK: Efficiency Robustness of Dynamic Deep Learning Systems at USENIX Security 2025 (Oral)
SoK: Efficiency Robustness of Dynamic Deep Learning Systems at USENIX Security 2025 (Poster)
CodeImprove: Program Adaptation for Deep Code Models at ICSE 2025 (Oral)
An Investigation on Numerical Bugs in GPU Programs Towards Automated Bug Detection at ISSTA 2025 (Contributed Presentation)
On the Fly Input Refinement for Deep Learning Models at the Doctoral Symposium ICSE 2025 (Oral)
On the Fly Input Refinement for Code Language Models at the Student Research Competition at ICSE 2025 (Poster)
Exploiting Efficiency Vulnerabilities in Dynamic Deep Learning Systems at Euro S&P 2025 (Virtual Poster)
Can you mimic me? Exploring the Use of Android Record & Replay Tools in Debugging at MOBILE-Soft 2025 (Oral)
Evaluation of Computer Performance using Cache Memory, Branch Prediction, and Pipelining at the Undergraduate Research Opportunities and Summer Workshop 2018, MSU Texas (Oral).
A study on Hardware Prefetching Effects in the Execution of Computer Programs at the Undergraduate Research Opportunities and Summer Workshop 2019, MSU Texas (Oral)
Multi-GPU Programming Pros and Cons: A Case Study at the Undergraduate Research Opportunities and Summer Workshop 2020, MSU Texas (Virtual Oral)
Performance Analysis of Machine Cycles for Good and Bad at the EURECA 2020, MSU Texas (Oral)

Guest Lectures and Talks

On Device Machine Learning, CS6375 (Machine Learning), The University of Texas at Dallas, Spring 2025
Efficiency Robustness of Dynamic Deep Learning Systems, Computer Science Faculty Mixture organized by The University of Texas at Dallas, Spring 2025
On the Fly Input Adaptation for DL/LLM-based Systems, Computer Science Faculty Mixture organized by The University of Texas at Dallas, Fall 2024
Automated Numerical Bug Detection for GPU Programs, Computer Science Faculty Mixture organized by The University of Texas at Dallas, Spring 2024
Empirical Investigation on Automated Numerical Bug Detection for GPU Programs, Computer Science Faculty Mixture organized by The University of Texas at Dallas, Spring 2023
Empirical Investigation on Automated Numerical Bug Detection for GPU Programs, Computer Science Faculty Mixture organized by The University of Texas at Dallas, Fall 2022
An Introduction to Microservices, CS3354 (Software Engineering), The University of Texas at Dallas, Summer 2021
Low-Code/No-Code Development, CS3354 (Software Engineering), The University of Texas at Dallas, Summer 2021

Page updated

Google Sites

Report abuse