Publications

SealQA introduces a challenging benchmark for evaluating search‑augmented language models on fact‑seeking questions, confronting noisy, conflicting, or misleading web search results. It includes three variants—Seal‑0 (core set), Seal‑Hard, and LongSeal (multi‑document with distractors)—and reveals major performance gaps: current frontier LLMs, even with search tools, achieve only ~17% accuracy on Seal‑0, show no consistent gains from more compute, and struggle with document retrieval in LongSeal. The dataset is publicly released to encourage progress in robust, retrieval‑augmented reasoning.

📋

@misc{pham2025sealqa,
  title={SealQA: Raising the Bar for Reasoning in Search-Augmented Language Models},
  author={Thinh Pham and Nguyen Nguyen and Pratibha Zunjare and Weiyuan Chen and Yu-Min Tseng and Tu Vu},
  year={2025},
  eprint={2506.01062},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

—This paper addresses the challenge of transforming complex sentences into sequences of logical, simplified sentences while preserving semantic and logical integrity with the help of Large Language Models. We propose a hybrid approach that combines advanced prompting with multi-agent architectures to enhance the sentence simplification process. Experimental results show that our approach was able to successfully simplify 70% of the complex sentences written for video game design application. In comparison, a single-agent approach attained a 48% success rate on the same task.{index=1}.

📋

@misc{zunjare2025hybrid,
  title={A Hybrid Multi-Agent Prompting Approach for Simplifying Complex Sentences},
  author={Zunjare, Pratibha and Hsiao, Michael S.},
  year={2025},
  eprint={2506.11681},
  archivePrefix={arXiv},
  primaryClass={cs.CL}
}

Real-time distributed systems (RTDS) face unpredictable changes such as machine failures, load spikes, or estimation errors, which can degrade performance. This book presents algorithms and techniques for robust resource allocation under such uncertainties, ensuring stability and controlled degradation of performance metrics like schedule length, load balance, and queue waiting time by accounting for estimation variability in resource demands.

📋

@book{zunjare2012robust,
  title={Robust Resource Allocation Algorithms in Real‑Time Distributed System: Robust Resource allocation Techniques},
  author={Zunjare, Pratibha and Sahoo, Bibhudatta},
  year={2012},
  publisher={LAP Lambert Academic Publishing},
  isbn={9783847378129},
  pages={92}
}

Real-time systems operate in unpredictable, changing environments that can degrade performance. This paper evaluates the robustness of resource allocation under Earliest Deadline First (EDF) scheduling on uniprocessor systems, using makespan as the performance metric. It defines deterministic and stochastic robustness metrics, introduces the FePIA procedure and experiments showing EDF’s robustness scales with task volume, and compares EDF vs. RM scheduling under varying loads.

📋

@article{zunjare2012robustness,
  title={Evaluating Robustness of Resource Allocation in Uniprocessor Real-Time Systems},
  author={Pratibha Zunjare and Bibhudatta Sahoo},
  journal={International Journal of Computer Applications},
  volume={40},
  number={3},
  pages={13--19},
  year={2012},
  publisher={IJCA}
}