Raghavendra Addanki

I am a Research Scientist at Adobe Research (San Jose, California). I completed my Ph.D. in Computer Science from the College of Information and Computer Sciences at the University of Massachusetts Amherst in May 2022, where I was co-advised by Prof. Andrew McGregor and Prof. Cameron Musco. I received a Dissertation Writing Fellowship for my Ph.D. thesis. I was a visiting student in the Causality program at the Simons Institute for the Theory of Computing, Berkeley in 2022. Prior to this, I completed my Bachelors and Masters degrees (dual degree) in Computer Science from the Indian Institute of Technology Madras in 2016.

Contact Email: raddanki AT adobe.com

Google Scholar / LinkedIn / Github / Twitter

Research Interests

I'm interested in understanding and building foundations of applications of generative models by developing and borrowing ideas from theoretical computer science, causal inference, reinforcement learning, and statistics.

Lately I have been interested in solving it with a full stack approach:

1. Agent harnesses: designing configurable, event- and time-driven harnesses that orchestrate agents across long-running tasks.

2. Building tools: building tools for data science and root cause analysis that operate on massive-scale ingested data, where the tools and agents are co-designed and optimized together.

3. Evaluation: evaluation is rapidly evolving from single LLM calls to full systems; designing metrics that isolate individual components while still evaluating the end-to-end system effectively.

Some time ago, during my Ph.D., among other things, I have worked on identifying new connections between combinatorial optimization and causal inference.

Professional Activities

Co-organizing the Workshop on Systems for Data-centric Agents with Human-in-the-loop (DASHSys), colocated with VLDB 2026. Register to participate in the competition.

Internships

I'm extremely lucky to have worked with the following interns: Yaswanth Chittepu, Noah Amsel, Shivanshu Shekhar, Shripad Deshmukh, Vignesh Viswanathan, Lei Shi, Andrei Graur, An Yan, among others.

Publications

Author ordering for the papers below is alphabetical unless marked *

CAPO: Counterfactual Credit Assignment in Sequential Cooperative Teams*
Shripad Deshmukh, Jayakumar Subramanian, Raghavendra Addanki, Nikos Vlassis
In submission (Conference on Neural Information Processing Systems (NeurIPS) 2026)
arXiv

ML-Tool-Bench: Tool-Augmented Planning for ML Tasks*
Yaswanth Chittepu, Raghavendra Addanki, Tung Mai, Anup Rao, Branislav Kveton
In submission (Conference on Language Modeling (COLM) 2026)
arXiv / pdf

Stationarity-Aware Causal Discovery in Time Series via Minimal Separating Sets*
Shanyun Gao, Raghavendra Addanki, Tong Yu, Ryan A. Rossi, Qifan Song, Murat Kocaoglu
International Conference on Artificial Intelligence and Statistics (AISTATS), 2026
pdf

GT-SVJ: Generative-Transformer-Based Self-Supervised Video Judge For Efficient Video Reward Modeling*
Shivanshu Shekhar, Uttaran Bhattacharya, Raghavendra Addanki, Mehrab Tanjim, Somdeb Sarkhel, Tong Zhang
IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2026
arXiv

Offline RL by Reward-Weighted Fine-Tuning for Conversation Optimization*
Subhojyoti Mukherjee, Viet Dac Lai, Raghavendra Addanki, Ryan A. Rossi, Seunghyun Yoon, Trung Bui, Anup Rao, Jayakumar Subramanian, Branislav Kveton
Conference on Neural Information Processing Systems (NeurIPS), 2025
arXiv

Leveraging semantic similarity for experimentation with AI-generated treatments*
Lei Shi, David Arbour, Raghavendra Addanki, Ritwik Sinha, Avi Feller
Conference on Neural Information Processing Systems (NeurIPS), 2025
arXiv / code

Traceable and Explainable Multimodal Large Language Models: An Information-Theoretic View*
Zihan Huang, Junda Wu, Rohan Surana, Raghav Jain, Tong Yu, Raghavendra Addanki, David Arbour, Sungchul Kim, Julian McAuley
Conference on Language Modeling (COLM), 2025
proceedings

Causal Discovery-Driven Change Point Detection in Time Series*
Shanyun Gao, Raghavendra Addanki, Tong Yu, Ryan A. Rossi, and Murat Kocaoglu
International Conference on Artificial Intelligence and Statistics (AISTATS), 2025
arXiv / proceedings / code

EigenLoRA: Recycle trained Adapters for Resource Efficient Adaptation and Inference*
Prakhar Kaushik, Aayush Mishra, Ankit Vaidya, Raghavendra Addanki, Ryan A. Rossi, Ani Nenkova, Anqi Liu, Alan Yuille, and Jiuxiang Gu
pdf

One Pass Streaming Algorithm for Super Long Token Attention Approximation in Sublinear Space
Raghavendra Addanki, Chenyang Li, Zhao Song, Chiwun Yang
arXiv

Parameter-Efficient Fine-Tuning via Partially Decomposable Loss Analysis and Sharing
Raghavendra Addanki, Ritwik Sinha, Zhao Song, Yizhou Wang, Lichen Zhang
pdf

Limits of Approximating the Median Treatment Effect
Raghavendra Addanki and Siddharth Bhandari
Conference on Learning Theory (COLT), 2024
arXiv / proceedings

Continuous Treatment Effects with Surrogate Outcomes*
Zhenghao Zeng, David Arbour, Avi Feller, Raghavendra Addanki, Ryan Rossi, Ritwik Sinha, and Edward H. Kennedy
International Conference on Machine Learning (ICML), 2024
arXiv / proceedings

Causal Discovery in Semi-Stationary Time Series*
Shanyun Gao, Raghavendra Addanki, Tong Yu, Ryan A. Rossi, and Murat Kocaoglu
Neural Information Processing Systems (NeurIPS), 2023
arXiv / proceedings / code

Anytime-Valid Confidence Sequences in an Enterprise A/B Testing Platform*
Akash Maharaj, Ritwik Sinha, David Arbour, Ian Waudby-Smith, Simon Liu, Moumita Sinha, Raghavendra Addanki, Aaditya Ramdas, Manas Garg and Viswanathan Swaminathan
ACM Web Conference (WWW), Industry Track, 2023
arXiv / proceedings

Sample Constrained Treatment Effect Estimation
Raghavendra Addanki, David Arbour, Tung Mai, Cameron Musco, and Anup Rao
Conference on Neural Information Processing Systems (NeurIPS), 2022
arXiv / proceedings / code

Non-Adaptive Edge Counting and Sampling via Bipartite Independent Set Queries
Raghavendra Addanki, Andrew McGregor, and Cameron Musco
European Symposium on Algorithms (ESA), 2022
arXiv / proceedings

Improved Approximation and Scalability for Fair Max-Min Diversification
Raghavendra Addanki, Andrew McGregor, Alexandra Meliou, and Zafeiria Moumoulidou
International Conference on Database Theory (ICDT), 2022
arXiv / proceedings

Collaborative Causal Discovery with Atomic Interventions
Raghavendra Addanki, Shiva Prasad Kasiviswanathan
Conference on Neural Information Processing Systems (NeurIPS), 2021
arXiv / proceedings / 12 min video at NeurIPS

Intervention Efficient Algorithms for Approximate Learning of Causal Graphs
Raghavendra Addanki, Andrew McGregor, and Cameron Musco
International Conference on Algorithmic Learning Theory (ALT), 2021
arXiv / proceedings / 1hr video at MIT / 12 min video at ALT

How to Design Robust Algorithms using Noisy Comparison Oracle
Raghavendra Addanki, Sainyam Galhotra, and Barna Saha
International Conference on Very Large Data Bases (VLDB), 2021
arXiv / proceedings

Efficient Intervention Design for Causal Discovery with Latents
Raghavendra Addanki, Shiva Prasad Kasiviswanathan, Andrew McGregor, and Cameron Musco
International Conference on Machine Learning (ICML), 2020
arXiv / proceedings / 15 min video at ICML / 1 hr video at NUS

Search Result Diversification with Guarantee of Topic Proportionality*
Sheikh Muhammad Sarwar, Raghavendra Addanki, Ali Montazeralghaem, Soumyabrata Pal, and James Allan
International Conference on the Theory of Information Retrieval (ICTIR), 2020
proceedings

Dynamic Set Cover: Improved Algorithms and Lower Bounds
Amir Abboud, Raghavendra Addanki, Fabrizio Grandoni, Debmalya Panigrahi, and Barna Saha
Symposium on Theory of Computing (STOC), 2019
proceedings

Embed as you need: Evaluation of Random Walk and Poincare Embeddings for Healthcare Tasks*
Khushbu Agarwal, Tome Eftimov, Raghavendra Addanki, Sutanay Choudhury, Suzanne Tamang, and Robert Rallo
Workshop on Applied Data Science for Healthcare, Knowledge Discovery and Data Mining (KDD), 2019
arXiv

Miscellaneous

Theoretical Computer Science group at UMass Amherst. There is a weekly theory seminar every semester -- if you are interested in giving a talk, please let one of the organizers know.

Website template borrowed from Jon Barron