Profile banner
Clark Peng

Clark Peng

UCLA CS undergraduate researching generative AI and video

Los Angeles, California
ucla+1
Joined May 2026
Favicon imageclarkipeng.github.io
Favicon imagescholar.google.com
Favicon imageorcid.org
Favicon imagelinkedin.com
Favicon imagehuggingface.co

Network

353 connections
🎓
UCLA CS Engineering Students
🌱
Incoming Tech Talent
🤖
AI ML Researchers
🎨
Product Design Professionals
🚀
Startup Founders
🏫
Sage Ridge School Staff

Summary

Researcher focused on generative models and robustness: Clark contributes to benchmark and evaluation efforts that probe limits of modern generative systems — including dialect robustness in image/video generation (DialectGen) and physical commonsense in video generation (VideoPhy-2). github+2
Active open-source contributor and developer: Clark has code and engineering contributions to projects such as TSMRealTime (real-time audio time-scale modification) and participates in shared GitHub projects used for research and benchmarking. github+1
UCLA undergraduate with campus leadership and teaching experience: As a UCLA Computer Science student, Clark combines research with student leadership (ACM AI president, previous projects officer) and instructional roles such as TA positions. github+1
Collaborator on large, multi-author benchmark projects: Clark participates in large collaborative efforts (e.g., Humanity's Last Exam, HLE) and co-authors multi-institution publications assessing AI capabilities. arxiv+1

Work

Education

Projects

Writing

VideoPhy-2: A Challenging Action-Centric Physical Commonsense Evaluation in Video Generation

January 1, 2025

Introduces VideoPhy-2, an action-centric dataset and evaluation framework to assess physical commonsense and semantic adherence in text-to-video generation; includes human evaluation and an automatic evaluator.

Favicon imagearxiv.org

DialectGen: Benchmarking and Improving Dialect Robustness in Multimodal Generation

January 1, 2025

Presents DialectGen, a large-scale multi-dialect benchmark for image and video generation, analysis of model failures on dialectal inputs, and an encoder-based mitigation approach.

Favicon imagearxiv.org

Humanity's Last Exam

January 1, 2025

Contribution to the large multi-author HLE benchmark project; HLE is a multi-modal expert-level academic benchmark intended to probe advanced AI capabilities.

Favicon imagearxiv.org

Event detection via probability density function regression

January 1, 2024

A time-series / event-detection preprint authored/co-authored by Clark Peng.

Favicon imagescholar.google.com

Hobbies

Enjoys competitive programming, game development, and sports. github