Naman Jain, Skanda Vaidyanath, Arun Iyer, Nagarajan Natarajan, Suresh Parthasarathy, Sriram Rajamani, Rahul Sharma
International Conference on Software Engineering (ICSE) 2022
Large pre-trained language models such as GPT-3, Codex, and Google's language model are now capable of generating code from natural language specifications of programmer intent. We view these developments with a mixture of optimism and caution. We present an approach to augment these large language models with post-processing steps based on program analysis and synthesis techniques, that understand the syntax and semantics of programs. Further, we show that such techniques can make use of user feedback and improve with usage.
Ramakrishna Bairi, Atharv Sonwane, Aditya Kanade, Vageesh D C, Arun Iyer, Suresh Parthasarathy, Sriram Rajamani, B. Ashok, Shashank Shet
Proceedings of the ACM on Software Engineering, Volume 1, Issue FSE, Article 31
Software engineering activities such as package migration, fixing error reports from static analysis or testing, and adding type annotations or other specifications to a codebase, involve pervasively editing the entire repository of code. We frame repository-level coding as a planning problem and present a task-agnostic, neuro-symbolic framework called CodePlan. CodePlan synthesizes a multi-step chain-of-edits (plan), where each step results in a call to an LLM on a code location with context derived from the entire repository, previous code changes and task-specific instructions.
Rahul Ragesh, Sundararajan Sellamanickam, Arun Iyer, Ramakrishna Bairi, Vijay Lingam
Proceedings of the 14th ACM International Conference on Web Search and Data Mining (WSDM '21)
We consider the problem of learning efficient and inductive graph convolutional networks for text classification with a large number of examples and features. We propose a heterogeneous graph convolutional network (HeteGCN) modeling approach that unites the best aspects of PTE and TextGCN together. The main idea is to learn feature embeddings and derive document embeddings using a HeteGCN architecture with different graphs used across layers, reducing model parameters significantly and enabling faster training.
Arun Iyer, Saketha Nath, Sunita Sarawagi
Proceedings of the 31st International Conference on Machine Learning (ICML 2014), Pages 530-538
We investigate the use of maximum mean discrepancy (MMD) in a reproducing kernel Hilbert space (RKHS) for estimating class ratios in unlabeled collections. Our analysis establishes that the estimate is statistically consistent and provides an upper bound on the error in terms of intuitive geometric quantities like class separation and data spread. We propose a novel convex formulation that automatically learns the kernel to be employed in the MMD-based estimation.
Namit Katariya, Arun Iyer, Sunita Sarawagi
2012 IEEE 12th International Conference on Data Mining (ICDM), Pages 329-338
We estimate the accuracy of a classifier on a large unlabeled dataset based on a small labeled set and a human labeler. We develop a novel strategy of learning r bit hash functions to preserve similarity in accuracy values and show that our algorithm provides better accuracy estimates than existing methods. Experiments show 15% to 62% relative reduction in error compared to existing approaches, while reading three orders of magnitude less data.
Pritish Chakraborty, Aditya Singh, Indradyumna Roy, Lokesh Nagalapatti, Himanshu Dutta, Arun Iyer, Yashoteja Prabhu, Gaurav Sinha, Manik Varma, Soumen Chakrabarti, Abir De
Advances in Neural Information Processing Systems (NeurIPS 2026) — Accepted as a Poster
Neural subgraph retrieval models degrade under query distribution shifts, while collecting fresh training labels requires expensive subgraph-matching computations. We introduce ACROSS, an active corpus selection framework that selects a small, label-efficient corpus subset for each out-of-distribution query under a fixed solver budget. ACROSS combines linearized corpus embeddings, randomized projections, and perturbation-tolerant representation learning to build a reusable corpus index. Experiments on four molecular graph datasets demonstrate improved downstream retrieval accuracy over six baselines.
Naman Gupta, Vaibhav Singh, Arun Iyer, Kirankumar Shiragur, Pratham Grover, Ramakrishna B. Bairi, Ritabrata Maiti, Sankarshan Damle, Shachee Mishra Gupta, Rishikesh Maurya, Vageesh D. C.
ICLR 2026 Workshop on Memory for LLM-Based Agentic Systems (MemAgents)
Sequential multi-agent reasoning frameworks such as Chain-of-Agents (CoA) handle long-context queries by decomposing inputs into chunks and processing them sequentially using LLM-based worker agents that read from and update a bounded shared memory. In this work, we study the problem of chunk ordering for long-context reasoning. We use the well-known Chow-Liu trees to learn a dependency structure that prioritizes strongly related chunks. Empirically, we show that a breadth-first traversal of the resulting tree yields chunk orderings that reduce information loss across agents and consistently outperform both default document-chunk ordering and semantic score-based ordering.
Divyanshu Saxena, Rishikesh Maurya, Xiaoxuan Ou, Gagan Somashekar, Shachee Mishra Gupta, Arun Iyer, Yu Kang, Chetan Bansal, Aditya Akella, Saravan Rajmohan
ICSE 2026 Workshop on Large Language Models for Code
The rapid adoption of AI agents across domains has made systematic evaluation crucial for ensuring their usefulness and successful production deployment. We propose a process of benchmark generation that helps evolve the benchmarks as the requirements change and perform robust evaluation of evolving AI agents. We instantiate this approach for a case study of service migration from one deployment platform to another at a large public enterprise.
Shashank Kirtania, Naman Gupta, Priyanshu Gupta, Sumit Gulwani, Arun Iyer, Suresh Parthasarathy Iyengar, Arjun Radhakrishna, Sriram K. Rajamani, Gustavo Soares
Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track (EMNLP 2025)
Large Language Models (LLMs) often generate incorrect or outdated information, especially in low-resource settings or when dealing with private data. To address this, Retrieval-Augmented Generation (RAG) uses external knowledge bases (KBs), but these can also suffer from inaccuracies. We introduce STACKFEED, a novel Structured Textual Actor-Critic Knowledge base editing with Feedback approach that iteratively refines the KB based on expert feedback using a multi-actor, centralized critic reinforcement learning framework.
Naman Gupta, Shreeyash Gowaikar, Arun Iyer, Kirankumar Shiragur, Ramakrishna B. Bairi, Rishikesh Maurya, Ritabrata Maiti, Sankarshan Damle, Shachee Mishra Gupta
First Workshop on Foundations of Reasoning in Language Models (FoRLM) at NeurIPS 2025
Reasoning over very long inputs remains difficult for large language models (LLMs). We introduce COSMIR (Chain Orchestrated Structured Memory for Iterative Reasoning), a chain-style framework that replaces ad hoc messages with a structured memory. A Planner agent first turns a user query into concrete, checkable sub-questions. Worker agents process chunks via a fixed micro-cycle: Extract, Infer, Refine, writing all updates to the shared memory. A Manager agent then Synthesizes the final answer directly from the memory.