Parsa Hosseini
PhD Student in Computer Science · University of Maryland
About Me
I am a PhD student in Computer Science at the University of Maryland, where I am advised by Soheil Feizi in the Reliable AI Lab. My research focuses on post-training large language models, with an emphasis on reasoning efficiency, reinforcement learning for reasoning and alignment, and reducing the cost of long-form reasoning. I also study failure modes of multimodal LLMs and tool use in agentic AI, with the broader goal of understanding and improving the reliability of modern AI systems. Before my PhD, I received my Bachelor's degree in Computer Engineering from Sharif University of Technology.
In summer 2026 I am a research intern at Capital One in New York, working on efficient reasoning and LLM post-training.
Outside of research, you'll usually find me doing calisthenics or weight lifting.
News
- Sep 2026 New preprint: "Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency" — arXiv
- Sep 2026 Heading to COLM 2026 in San Francisco this October to present CoDE-Stop at the Workshop on Efficient Reasoning — arXiv
- Jul 2026 Our paper "Your LLM Agents are Temporally Blind" was accepted to ACL 2026 Findings — arXiv
- Jun 2026 Started a research internship at Capital One in New York, working on efficient reasoning and LLM post-training
- May 2026 Received my MSc in Computer Science along the way at the University of Maryland with a 4.0/4.0 GPA 🎓
- Apr 2026 New preprint: "Early Stopping for Large Reasoning Models via Confidence Dynamics" (CoDE-Stop) — arXiv
- Jan 2026 "GHOST: Hallucination-Inducing Image Generation for Multimodal LLMs" was accepted at ICLR 2026 — arXiv
- 2025 "SpurLens: Automatic Detection of Spurious Cues in Multimodal LLMs" was accepted to TMLR — arXiv
- Aug 2025 "Gaming Tool Preferences in Agentic LLMs" was accepted to EMNLP 2025 (Main Track) — arXiv
- Dec 2024 Co-organizing the Workshop on Spurious Correlation and Shortcut Learning at ICLR 2025
- Aug 2024 Started my PhD in Computer Science at the University of Maryland 🎓
- Jun 2024 "Decompose-and-Compose" was presented at CVPR 2024 — arXiv
Publications
* denotes equal contribution
Experience
Education
Awards & Service
- Dean's Fellowship, Dept. of Computer Science, University of Maryland (2024, 2025)
- Gold medal, Iranian Mathematical Olympiad (2018)
- Silver medal, Iranian Geometry Olympiad (2018)
- Co-organizer, Workshop on Spurious Correlation and Shortcut Learning @ ICLR 2025
- Reviewer: NeurIPS, ICLR, ICML, ACL, COLM