Elaine Wan's Research at UCLA: Advancing Trustworthy AI and Multimodal Generative Models

Introduction

Elaine Wan is a Computer Science PhD student at UCLA, advised by Professor Kai-Wei Chang, and a member of the UCLANLP research group. Her research focuses on building trustworthy multimodal generative models, specifically improving controllability, fairness, and factuality in text and image generation. Wan's work spans various aspects of AI, including natural language processing, computer vision, and human-computer interaction, with a strong emphasis on addressing biases and enhancing the utility of AI systems for diverse users.

Academic Background and Affiliations

Elaine Wan holds a B.S. in Applied Mathematics (with a double major in Economics) from UCLA. Her academic journey reflects a multidisciplinary approach, combining mathematical rigor with economic principles, which informs her current research in AI. She is currently a research intern at Tencent AI Lab in Bellevue, WA, working on controllable image editing under the mentorship of Lei Ke and Wenhao Yu. Previously, she interned at Amazon AGI and Microsoft Research Asia (MSRA), gaining experience in cutting-edge AI research and development.

Research Focus: Trustworthy Multimodal Generative Models

Wan's research is centered around building trustworthy multimodal generative models. This involves creating AI systems that can generate content across different modalities, such as text and images, while ensuring that the generated content is controllable, fair, and factual. Her work addresses critical challenges in AI, including:

• Controllability: Developing methods to allow users to have greater control over the content generated by AI models, enabling them to specify desired attributes and characteristics.
• Fairness: Identifying and mitigating biases in AI models to ensure that they do not perpetuate or amplify societal stereotypes and prejudices.
• Factuality: Ensuring that the information generated by AI models is accurate and consistent with real-world knowledge.

Key Publications and Projects

Elaine Wan's research has resulted in several notable publications and projects, showcasing her contributions to the field of AI. Some of her key works include:

1. MotionEdit: Benchmarking and Learning Motion-Centric Image Editing: This project focuses on motion image editing, introducing a benchmark and learning framework for manipulating motion in images. The work was accepted to CVPR 2026 and highlights Wan's expertise in computer vision and generative models.

   Read also: UCLA vs. Illinois: Basketball History

2. LLM Unlearning and Fairness: Wan has contributed to research on unlearning in large language models (LLMs) and fairness in AI systems. Her papers on this topic, accepted to EMNLP 2025 Findings, include:

   • Not Every Token Needs Forgetting: Selective Unlearning to Limit Change in Utility in Large Language Model Unlearning: This paper explores selective unlearning techniques to minimize the impact on model utility when removing specific information from LLMs.
   • LUME: LLM Unlearning with Multitask Evaluations: This work introduces a multitask evaluation framework for assessing the effectiveness of LLM unlearning methods.
   • Fact-or-Fair: A Checklist for Behavioral Testing of AI Models on Fairness-Related Queries: This paper proposes a checklist for evaluating the fairness of AI models by testing their behavior on fairness-related queries.

3. Fairness in Text-To-Image Generation: Wan's research also addresses fairness in text-to-image generation, with papers accepted to ACL 2025 Main:

   • The Male CEO and the Female Assistant: Evaluation and Mitigation of Gender Biases in Text-To-Image Generation of Dual Subjects: This paper examines gender biases in text-to-image generation models and proposes methods for mitigating these biases.
   • White Men Lead, Black Women Help? Benchmarking and Mitigating Language Agency Social Biases in LLMs: This work benchmarks and mitigates language agency social biases in LLMs, aiming to promote more equitable and representative content generation.

4. "Kelly is a Warm Person, Joseph is a Role Model": Gender Biases in LLM-Generated Reference Letters: This research investigates gender biases in reference letters generated by large language models.

5. Theoremqa: A theorem-driven question answering dataset: Wan contributed to the creation of a dataset for theorem-driven question answering.

6. Survey of Bias In Text-to-Image Generation: Definition, Evaluation, and Mitigation: This survey paper provides a comprehensive overview of bias in text-to-image generation, covering definitions, evaluation methods, and mitigation techniques.

   Read also: Navigating Tech Breadth at UCLA

7. Are personalized stochastic parrots more dangerous?: This work explores the potential risks associated with personalized stochastic parrots.

Contributions to Human-Computer Interaction and Accessibility

Wan's research extends to the field of human-computer interaction (HCI), with a focus on improving accessibility and inclusivity. Her work in this area includes:

• Enriching YouTube Experiences for Blind and Low-Vision Audiences: Developing a system that encourages viewers to post structured descriptive comments on YouTube videos, enhancing the experience for blind and low-vision users.
• Human-AI Collaborative Annotation System for People with Limited Mobility: Creating a system that uses an LLM-based approach to learn each user’s mobility profile and identify personalized accessibility barriers from everyday photos. This research also involves longitudinal user studies to assess the effectiveness of personalized LLMs in assisting people with disabilities.
• Personalized Camera Masking for Privacy: Introducing a privacy mechanism that uses a smartwatch as a signal guide to enable real-time, personalized camera masking, allowing users to obfuscate privacy-sensitive pixels while preserving the utility of AI vision.
• Interaction-Powered LEDs for Sensing and Communication: Developing "interaction-powered LEDs" that enable ordinary objects to emit identifiable light patterns for sensing and communication without the need for batteries or chips.

Awards and Recognition

Elaine Wan's research has been recognized with several awards and honors, including the 2025 Amazon AI Fellowship. This prestigious fellowship acknowledges her exceptional contributions to the field of AI and supports her ongoing research efforts.

UCLA's Research Ecosystem

Elaine Wan's work is supported by UCLA's vibrant research ecosystem, which includes various centers, labs, and initiatives focused on advancing AI and related fields. Some notable entities at UCLA include:

• UCLANLP Research Group: A leading research group in natural language processing, providing a collaborative environment for Wan's research.
• UCLA Center for Heterogeneous Integration and Performance Scaling (CHIPS): This center focuses on advancing heterogeneous integration and performance scaling in computing systems.
• Ozcan Lab: Known for its work on optical generative models, as highlighted in a Nature paper covered by IEEE Spectrum.
• UCLA Integrated Sensors Laboratory: Led by Prof. Aydin Babakhani, this lab focuses on developing integrated sensors and systems, including broadband receivers for various applications.
• UCLA Amazon Science Hub: Facilitates collaboration between UCLA researchers and Amazon, supporting research in AI and related areas.

Opportunities at UCLA

UCLA offers numerous opportunities for students and researchers interested in AI and related fields. These opportunities include:

Read also: Understanding UCLA Counselors

• Research Courses: UCLA offers research courses such as SRP 199, 198, URSP, and URFP, providing students with hands-on research experience.
• Summer Programs: Programs like Amgen Scholars provide summer research opportunities for undergraduate students.
• Undergraduate Science Journal (USJ): UCLA’s premier student-run STEM research journal, offering a platform for students to publish their research.
• ECE Career Fair: Hosted by HKN, the ECE Career Fair connects students with leading companies in the electrical and computer engineering fields.
• Fellowship Programs: Programs like the Amazon AI Ph.D. Fellowship support graduate students pursuing research in AI.

External Collaborations and Opportunities

UCLA researchers also collaborate with external organizations and participate in various initiatives, expanding the impact of their work. Examples include:

• Tencent AI Lab: Elaine Wan's current internship at Tencent AI Lab provides her with valuable experience in industry research and development.
• Amazon AGI and Microsoft Research Asia (MSRA): Wan's previous internships at these organizations have broadened her research perspective and provided her with opportunities to work on cutting-edge AI projects.
• Center for International Security and Cooperation (CISAC) at Stanford University: This center offers fellowship programs for researchers interested in international security and cooperation.

Industry Connections and Recruitment

UCLA's ECE department has strong connections with industry, providing students and researchers with opportunities for internships, jobs, and collaborations. Companies that recruit at UCLA include:

• Cadence
• Qualcomm
• Marvell
• Infineon
• OSI Systems
• HRL Laboratories
• Aerospace Corporation
• LADWP
• Analog Bits
• Spectrant LLC
• RFA Electric
• Apple
• DMF Lighting

Additionally, deep-tech startups in Pasadena are seeking Electrical/Computer Engineering Interns to develop AI-powered product prototypes for space applications.

tags: #elaine #wan #ucla #research

Popular posts:

• Learn to Play: A Comprehensive Guide
• Understanding CDF in Education
• Comprehensive Guide to PwC Start Internship
• The role of SSDP in drug policy reform
• Deconstructing stereotypes of hot college women