Te-Lin Wu (吳德霖)
Research Engineer
Email: telinwu [at] cs (dot) ucla (dot) edu
Google Scholar / LinkedIn / Github
I am a researcher at Character.ai working on large language models (LLMs) and model post-training for smarter and more amusing chatbots.
Prior to that, I obtained my PhD at UCLA PlusLab advised by Nanyun (Violet) Peng,
where my research focuses on multimodal models across NLP and computer vision.
I have also worked with Joseph J. Lim on reinforcement learning and vision for robotics topics.
Prior to PhD, I obtained my M.S. at Stanford University where I was advised by Silvio Savarese,
and I did my undergrad at National Tsing-Hua University (國立清華大學).
Over the summers, I've been lucky to work as research interns in several wonderful groups, including
Google Research, Meta Reality Labs, Amazon AI, and Adobe Research.
Agent-DocEdit: Language-Instructed LLM Agent for Content-Rich Document Editing
COLM 2024 /
Paper
A modularized LLM-agent approach for content-rich multimodal document editing.
DACO: Towards Application-Driven and Comprehensive Data Analysis via Code Generation
NeurIPS Datasets and Benchmarks Track 2024 /
Paper
A new dataset for complex data analysis tasks and newly proposed RLHF techniques for the tasks.
VDebugger: Harnessing Execution Feedback for Debugging Visual Programs
EMNLP 2024 (Findings) /
Paper
A debugging tool for visual program generator.
LegalDiscourse: Interpreting When Laws Apply and To Whom
NAACL 2024 /
Paper
A novel legal-related dataset that emphasizes on the discourse and the taxonomy of span-and-relation parsing.
COM2SENSE: A Commonsense Reasoning Benchmark with Complementary Sentences
A dataset for complementary commonsense reasoning collected via a model-in-the-loop gamified session.
MELINDA: A Multimodal Dataset for Biomedical Experiment Method Classification
A dataset for multimodal biomedical method classification.
LAMPRET: Layout-Aware Multimodal PreTraining for Document Understanding
ViGIL Workshop, NAACL 2021 /
Paper
A pre-training paradigm to exploit document layout to learn a document representation.
Program Guided Agent
ICLR 2020
A framework to programmatically control an RL-trained agent.
Demo2Vec: Reasoning Object Affordances from Online Videos
CVPR 2018
Learning to infer object affordance with a video demonstration of how to interact with objects.
Feedback Networks
CVPR 2017
A study of feedback mechanism of convolutional neural networks.