Visual Distribution Anchoring for Efficient Prompt Tuning

Pouya Parsa, Raoof Zare Moayedi, and Seongjin Choi. arXiv:2607.28967. Preprint, July 2026.

VDA turns a disjoint unlabeled target pool into class-specific visual prototypes and fuses them with frozen semantic classifiers. The primary matched result improves mean accuracy from 65.82% to 69.21% while requiring no target-side optimization or test-query access.

Explore the project page · Read the paper

Can the Cloud Drive? Infrastructure Feasibility of Offloading Autonomous Driving Across 5G and 6G

Pouya Parsa, Kawon Han, and Seongjin Choi. arXiv:2607.09045, 2026.

The study shows why near-term cloud VLA inference is often limited by GPU memory bandwidth—even after 5G/6G can carry the workload—and when shared cloud GPUs can reduce cost.

Explore the interactive article · Read the paper

Video-based Vehicle Surveillance in the Wild: License Plate, Make, and Model Recognition with Self-Reflective Vision-Language Models

Pouya Parsa, Keya Li, Kara M. Kockelman, and Seongjin Choi. arXiv:2508.01387. Accepted for presentation at TRB 2026.

• Designed and implemented a self-reflective vision-language model framework.
• Achieved a 20% performance improvement over CNN-based baselines on license plate, make, and model recognition tasks.

Where2Start: Leveraging Initial States for Robust and Sample-Efficient Reinforcement Learning

Pouya Parsa, Raoof Zare Moayedi, Mohammad Bornosi, and Mohammad Mahdi Bejani. arXiv:2311.15089.

• Proposed and implemented the Where2Start algorithm to select informative initial states.
• Achieved up to 8x improvement in sample efficiency compared to baseline reinforcement learning methods.

Departmental Fellowship Award

2026

Awarded a competitive fellowship in recognition of academic excellence and research potential.

CTS TRB Reimbursement Award

2026

Received competitive travel reimbursement to present research at TRB 2026.

University of Minnesota Twin Cities

Master of Science, 2024-2026

GPA: 3.7/4.00. Selected courses: Computer Vision and Generative AI.

Amirkabir University of Technology (Tehran Polytechnic)

Bachelor of Science, 2018-2022

GPA: 3.72/4.00. Thesis: Transformer-Based Deep Learning Method for Portfolio Optimization.

Research Assistant, University of Minnesota

Minneapolis, USA · Aug 2024-Present

Advisor: Dr. Seongjin Choi. Conducting research in machine learning with a focus on computer vision and generative AI.

Project Manager, XTON

Remote (Dubai) · Apr 2024-Jul 2024

• Managed infrastructure for a high-availability crypto clicker app serving 500k+ users.
• Implemented serverless architecture leveraging AWS Lambda and auto-scaling.
• Facilitated stakeholder communication to align expectations.

Machine Learning Engineer, ZebraCat.ai

Remote (Germany) · Sep 2022-Mar 2023

• Developed an embedding-based algorithm for efficient video retrieval.
• Developed a model to divide videos into semantic sections.
• Scraped more than 1M advertisement videos.

Data Science Intern, MCI (Hamrahe Aval)

Tehran, Iran · Jul 2022-Sep 2022

• Contributed to recommender systems at Iran's largest mobile operator.
• Built a fault-tolerant ETL pipeline for recommender data preparation.
• Implemented collaborative filtering delivering posts with less than 10 ms latency.

Machine Learning Engineer Intern, University of Zurich

Zurich, Switzerland · Jan 2022-Mar 2022

Developed software interfaces for data ingestion, preprocessing, machine learning, visualization, and report generation.