Tarun Suresh

I'm a third-year undergraduate student in the Computer Science Department at the University of Illinois, Urbana-Champaign advised by Prof. Sasa Misailovic, Prof. Gagandeep Singh, and Prof. Heng Ji.

Email  /  LinkedIn  /  Google Scholar

I am deeply interested in the intersection of deep learning, systems, and formal methods/programming languages. My research goal is to develop principled techniques that enhance the effectiveness, scalability, and reliability of AI systems, particularly in software development and safety-critical domains.

I'm currently researching:

- Language Models for Program Synthesis, Math, Logic, and Reasoning Tasks

- Embedding Models, Re-Ranking, and Retrieval-Augmented Generation

- Inference-time Scaling of LLMs

- Formal Verification and Robustness of Deep Neural Networks

We introduce CoRNStack, a large-scale, high-quality contrastive training dataset for code that spans multiple programming languages. We demonstrate that contrastive training of embedding models and LLM re-rankers using CoRNStack leads to state-of-the-art performance across a variety of code retrieval tasks. Notably, our lightweight retriever + re-ranker achieves state-of-the-art repoistory level bug localization on SWE-Bench over top automated software development frameworks.

IterGen is a framework that enables users to define and enforce semantic constraints into LLM output and efficiently backtrack by grammar symbols and re-sample during LLM generation until these constraints are satisfied. IterGen improves LLM-generated SQL accuracy by 18% and eliminates LLM privacy leakage.

We introduce a scalable Branch-and-Bound-based verifier RABBit for precisely verifying hyperproperties such as monotonicity and fairness defined over multiple executions of Deep Neural Networks.

We develop a method, called TAR, for building tamper-resistant safeguards into open-weight LLMs such that adversaries cannot remove the safeguards even after thousands of steps of fine-tuning. In extensive evaluations and red teaming analyses, we find that our method greatly improves tamper-resistance while preserving benign capabilities.

We present IRS, the first probabilistic approach for 5x faster robustness re-certification of Deep Neural Networks after model compression (pruning, quantization) or fine-tuning

We present the first study of the robustness of existing watermarking techniques on code generated by large language models and propose a parsing-based algorithm that easily removes these watermarks via semantic preserving transformations of the code.

SynCode is a novel framework for the grammar-guided generation of Large Language Models (LLMs) that is scalable to general-purpose programming languages and has soundness and completeness guarantees. SynCode reduces syntax errors by 96-100% for various languages (JSON, Python, Go) and enables 1.5x-10x faster LLM inference than existing approaches.

To address challenges from the risk of reward hacking and the complexity of reinforcement learning during preference-based reinforcement learning, we develop a novel two-step learning method called PRC. The high-level idea is to limit the reinforcement learning agent to optimize over a constrained action space that excludes out-of-distribution state-actions, which are unreliable and increase the complexity of the reinforcement learning problem.

We propose efficient deterministic formal verifiers to speed up DNN re-verification after pruning, quantization, or fine-tuning.

Template from here