To obtain the slides view password, please contact me.
Sidewalk Moments: Are Richer Representations Always More Human-Aligned? Evidence from City-Walk Videos
Date: Mar 19, 2026
Location: San Francisco
This presentation reports an independent research project comparing video, temporally averaged images, audio, and text as representations of urban engagement in city-walk footage. The paper is under review at Scientific Reports. Project page
Seeing HAI: Human Activities and Interactions on Sidewalks and Pedestrian Spaces through VLMs
Date: Oct 23, 2025
Location: Minneapolis
This presentation reports one methodological part of my independent dissertation research. It discusses why social groups are difficult to represent as conventional object categories and introduces the three-stage MINGLE pipeline for detecting relational groups in street imagery. Dissertation project · MINGLE
Exploring the Challenges of Humanity’s Last Exam
Date: Aug 20, 2025
Location: Sunnyvale, CA
This presentation introduces my analysis of the Humanity’s Last Exam (HLE) benchmark at Amazon AGI, a 2,500-question dataset spanning STEM, medicine, and humanities. I discussed the dataset’s quality issues, small-model performance, and reasoning behaviors. The work examines how current distilled models perform under strict evaluation and presents a framework for analyzing reasoning errors and recurring problem-solving strategies. The technical report is listed in my Publications section.
Seeing HAI: Human Activities and Interactions on Sidewalks and Pedestrian Spaces through VLMs
Date: May 22, 2025
Location: 10-401, MIT
This presentation introduced my independent doctoral research, Seeing HAI: Human Activities and Interactions on Sidewalks and Pedestrian Spaces through VLMs. The dissertation is distinct from Sidewalk Ballet, a separate City Form Lab collaboration that I lead at MIT.
The dissertation begins with a three-stage method for detecting social groups in sidewalk scenes. It then expands from grouping to a wider set of visible activities and interactions and develops Human Activity and Interaction indicators at the sidewalk-segment level. Dissertation project page
All members of my dissertation committee attended the colloquium: Prof. Andres Sevtsuk, Prof. Jinhua Zhao, and Prof. Lawrence Vale. The external reviewer was Prof. Albert Saiz.
Approved dissertation proposal (PDF) — the document is password-protected; please email me for the password.
Data Benchmark for Multi-modality models
Date: Aug 22, 2024
Location: Sunnyvale, CA
This presentation covers two benchmark projects from my Amazon AGI research internship. Both examine audio generation, pairing music notation or video with audio and comparing model outputs with human judgments and automatic measures. The technical reports are listed in my Publications section.
Urban Big Data in China
Date: May 8, 2024
Location: 4-370, Massachusetts Institute of Technology
This presentation explores my entrepreneurship journey, focusing on urban big data analytics in China. Specifically, I discuss four distinct types of data: open data, web-crawled data, purchased data, and geo-tagged visual data. For each category, I describe 2-3 datasets that I have worked with and detail specific research questions and case studies associated with them.
Sidewalk Ballet
Date: Mar 21-22, 2024
Location: School of Architecture and Planning, Massachusetts Institute of Technology
At the Spring 2024 HPI–MIT Workshop, the MIT City Form Lab and Hasso Plattner Institute teams presented early work from Sidewalk Ballet. The collaboration examined available computer-vision and vision-language models for detecting social activity on sidewalks and developed the initial research pipeline. I serve as the Sidewalk Ballet project lead at MIT. Project page
Learning from ground floors
Date: Sep 22-23, 2023
Location: School of Architecture and Planning, Massachusetts Institute of Technology
The 2023 Rez de Ville: In Transition seminar will delve into the evolving dynamics of urban ground floors amidst modern societal, economic, environmental, and technological shifts and explore the implications of these changes on diverse urban populations. Here is the talk list. A quick brief of the case can be found here.
Mapping with Image
Date: Jul 26, 2023
Location: School of Architecture and Planning, Massachusetts Institute of Technology
A brief of my previous research and work. Due to the time limit, mostly is about urban imagery analysis. It is the application for the research work at Senseable City Lab. The process was smooth and successful.
Date: Oct 17, 2019
Location: Zhengzhou, China
Date: Jul 25, 2019
Location: Jakarta, Indonesia
My team is invited as a representative of FCIC to participate in the project initiation meeting on “Improving the Use and Sharing of Geospatial Information to Promote Resilient and Sustainable Development.” This event, hosted by the United Nations Economic and Social Commission for Asia and the Pacific (UN-ESCAP) and organized by the Indonesian National Institute of Aeronautics and Space (LAPAN). During this meeting, we engaged in vibrant discussions with representatives from Cambodia, Mongolia, Thailand, Kyrgyzstan, Indonesia, and experts from research institutions such as The Chinese University of Hong Kong, the Remote Sensing Institute of the Chinese Academy of Sciences, and the City University of Macau.
Experience with urban data
Date: Oct 12, 2022
Location: School of Architecture and Planning, Massachusetts Institute of Technology
A brief of my previous research and work. It’s a 45-min talk, the talk includes my previous data-driven research and practice. The audience are all City Form Lab members.
Seeing City
Date: May 14, 2019
Location: World Bank Beijing Office
It’s an 30 mins talk about recent projects from CitoryTech team. The talk covered a series of urban data research, practice in our exploration of understanding our city with images.
Date: Mar 11, 2019
Location: Faculty of Architecture, Hong Kong University
It’s an 1-hour talk about recent projects from CitoryTech team. The talk covered a series of urban data research, practice in our exploration of understanding our city with images. I also introduced the development of crowd sourcing tools.
StreeTalk Pitch
Date: Nov, 2017
Location: Shanghai
This is a 10-minute business plan presentation for our original project, StreeTalk. We were honored to receive the seed award and seed funding for CitoryTech.
Additionally, this project played a pivotal role in securing our first order from Daimler, through which we sold our safety maps, encompassing over 300 cities in China.
National Population Flow Analysis Based on 1 billion LBSN data
Date: Sep 26, 2016
Location: Shenyang
A presentation on the study of national population movements, utilizing the Tencent LBSN dataset, was delivered. For the first time, it showcased an analysis of urban clusters and the regional framework across the whole country.
A brief of the project can be found here.
GTFS and AFC Data Analysis
Date: Aug, 2013
Location: New York
This is a 30-minute presentation
