💻 AI for Software Engineering

This page highlights ongoing and past research efforts connecting AI, formal methods, programming languages and software engineering at the reasoning and learning research group @ Georgia Tech led by Professor Vijay Ganesh.

Vijay Ganesh's Homepage

RLSF: Fine-tuning LLMs via Symbolic Feedback (2025)

Authors: Piyush Jha¹, Prithwish Jana¹, Pranavkrishna Suresh¹, Arnav Arora¹, Vijay Ganesh¹

Affiliation: ¹Georgia Institute of Technology, USA

TL;DR: A novel fine-tuning paradigm where symbolic reasoning tools (e.g., solvers, provers, RDKit) provide token-level feedback to LLMs, achieving large accuracy gains on code synthesis, chemistry, and math tasks while remaining far smaller than closed-source models.

Published in ECAI-2025 arXiv Project Page

CoTran: An LLM-based Code Translator using Reinforcement Learning with Feedback from Compiler and Symbolic Execution (2024)

Authors: Prithwish Jana¹, Piyush Jha¹, Haoyang Ju², Gautham Kishore³, Aryan Mahajan⁴, Vijay Ganesh¹

Affiliations: ¹Georgia Institute of Technology, USA | ²University of Toronto, Canada | ³UC San Diego, USA | ⁴Columbia University, USA

TL;DR: Fine-tunes LLMs for end-to-end code translation using compiler and symbolic-execution feedback to improve compilability and functional equivalence.

Published in ECAI-2024 arXiv GitHub

Grounding Neural Inference with Satisfiability Modulo Theories (2023)

Authors: Zifan Wang¹, Saranya Vijayakumar², Kaiji Lu³, Vijay Ganesh⁴, Somesh Jha⁵, Matt Fredriskon²

Affiliations: ¹Center for AI Safety | ²Carnegie Mellon University, USA | ³Pinterest Inc. | ⁴Georgia Institute of Technology, USA | ⁵University of Wisconsin, Madison, USA

TL;DR: Introduces SMTLayer, enabling neural networks to enforce logical constraints directly during both forward and backward passes without requiring the solver to be differentiable.

Published in NeurIPS-2023 GitHub

CGDTest: A Constrained Gradient Descent Algorithm for Testing Neural Networks (2023)

Authors: Vineel Nagisetty¹, Laura Graves¹, Guanting Pan¹, Piyush Jha¹, Vijay Ganesh¹

Affiliation: ¹University of Waterloo, Canada

TL;DR: Applies constrained gradient descent to systematically test deep networks, uncovering adversarial robustness and fairness issues more effectively than existing testing tools.

arXiv Talk

A Solver + Gradient Descent Training Algorithm for Deep Neural Networks (2022)

Authors: Dhananjay Ashok¹, Vineel Nagisetty², Christopher Srinivasa², Vijay Ganesh³

Affiliations: ¹University of Toronto, Canada | ²Borealis AI | ³University of Waterloo, Canada

TL;DR: Combines gradient descent with MILP solver steps to escape poor local minima and accelerate convergence on classification and regression tasks.

Published in IJCAI-2022 Project Page GitHub

Amnesiac Machine Learning (2021)

Authors: Laura Graves¹, Vineel Nagisetty¹, Vijay Ganesh¹

Affiliation: ¹University of Waterloo, Canada

TL;DR: Develops unlearning techniques that remove the influence of specific training data to satisfy privacy laws like GDPR while maintaining model performance.

Published in AAAI-2021 arXiv Project Page GitHub

xAI-GAN: Enhancing Generative Adversarial Networks via Explainable AI Systems (2021)

Authors: Vineel Nagisetty¹, Laura Graves¹, Joseph Scott¹, Vijay Ganesh¹

Affiliation: ¹University of Waterloo, Canada

TL;DR: Leverages explainability to guide GAN training, improving both the interpretability and quality of generated outputs.

Published in AAAI-21 (Explainable Agency Workshop) arXiv Project Page GitHub