Scholar
Aakriti Agrawal
Google Scholar ID: 2RRnCRMAAAAJ
University of Maryland, College Park
Citations & Impact
Citations
292
H-index
6
i10-index
6
Publications
14
Co-authors
18
list available
Publications
Resume (English only)
Academic Achievements
- Paper 'Uncertainty-Aware Answer Selection for Improved Reasoning in Multi-LLM Systems' accepted at EMNLP 2025
- 'EnsemW2S: Can an Ensemble of SoTA LLMs be Leveraged to Obtain a Stronger LLM?' accepted at NeurIPS 2024 SafeGenAI workshop; under review at AAAI 2026
- Co-authored 'Easy2Hard-Bench' published in NeurIPS 2024 Dataset Track
- 'Robustness to Multi-Modal Environment Uncertainty in MARL using Curriculum Learning' published at NeurIPS 2023 MASEC workshop
- Co-authored 'WAVES: Benchmarking the Robustness of Image Watermarks' published at ICML 2024
- Co-authored 'PoisonedParrot' accepted at NAACL 2025 and presented at SafeGenAI @ NeurIPS 2024
- 'Towards Mitigating Hallucinations in Large Vision-Language Models by Refining Textual Embeddings' under review
Background
- 4th-year Computer Science PhD student at the University of Maryland, College Park
- Research interests include: improving reasoning in LLMs, reducing reward hacking in reasoning LLMs
- Weak-to-strong generalization and self-improvement
- LLM/VLM alignment, reasoning, and hallucination reduction
- Reinforcement Learning, Multi-Agent Systems, and Uncertainty Estimation
