Interpretable Prediction of Urban Population Flow | Computational Demography Reading Group #7

Explainable Prediction of Urban Population Mobility | Computational Demography Book Club Episode 7

Introduction

Daily travel in cities provides a concrete entry point for computational demography to study population movement and its spatiotemporal patterns. In this 7th session of the Computational Demography Book Club, Meituan intelligent decision‑algorithm researcher Ji Jiahao uses urban traffic forecasting as an example to show how to combine crowd‑movement mechanisms with deep learning, exploring how to leverage urban mobility data to improve prediction accuracy while offering clearer explanations of mobility changes.

To systematically review the latest advances in the field of demography, Sanhe Lian Tong Technology and Jizhi Club jointly launched the “Computational Demography: Fertility, Education, Mobility, and Aging from a Complex Systems Perspective” reading group. Professors Li Handong and Cui Xuefeng from the School of Systems Science, Beijing Normal University, serve as co‑initiators. The group brings together researchers and explorers interested in this topic to study cutting‑edge literature and exchange research ideas.

The reading group will be held online every Thursday from 19:30 to 21:30 starting September 17, 2026, lasting about 10 weeks, including a main presentation and discussion, and will provide video recordings after the sessions. We sincerely invite researchers in related fields and those with interdisciplinary interests to participate.

Report Overview

Everyday travel in cities provides a concrete entry point for demographic research on population mobility and its spatiotemporal patterns. This talk uses urban traffic forecasting as an example to show how to combine mobility mechanisms with deep learning: describing inter‑area flows and their evolution on the road network with potential fields and continuity equations, handling environmental changes such as weather and holidays through causal modeling, and discussing from an information‑theoretic perspective when “decompose first, then predict” is more effective. Around these topics, we explore how to leverage urban mobility data to improve forecasting accuracy while offering clearer explanations of mobility changes.

Share Outline

Urban Population Mobility Forecast: Limits of Data‑Driven Approaches

• Observing urban mobility through traffic data

• Beyond prediction accuracy: interpretability and environmental adaptability

Potential fields and continuity equations: from regional flow to network evolution

• Describing inter‑regional flow with potential fields

• Characterizing dynamic evolution on road networks with continuity equations

• Predictive Performance and Explainability of Potential Fields

Causal Modeling: Robust Prediction under Environmental Changes

• Distribution shifts caused by weather, holidays, etc.

• Confounder representation based on basis vectors

• Predictive performance in unknown environments

Decomposed Forecasting: When “Decompose First, Predict Later” Is More Effective

• Modeling challenges from mixed travel modes

• Component independence principle from an information-theoretic perspective

• Graph Decomposition Learning Framework and Experimental Validation

Core Concepts

Spacetime Potential Field: Borrowing from field theory, using the “potential difference” between regions to describe flow direction and intensity.

Neural Differential Equations: combine differential equations with neural networks to describe the continuous evolution of traffic flow.

Causal Representation Learning: distinguish relatively stable spatiotemporal relationships from the influence of external environmental factors.

Decomposed Forecasting: split complex spatiotemporal patterns into relatively independent components, forecast each separately, and then recombine.

Speaker Introduction

Speaker: Ji Jiahao, Meituan Intelligent Decision Algorithm Researcher, Ph.D. from Beihang University, twice participated in overseas research through national government-sponsored programs. Research interests include AI4OR, spatiotemporal data mining, explainable AI, and urban computing. Serves as a member of the Federation Committee of the Chinese Association of Automation, co-authored the international standard IEEE P2894 on explainable artificial intelligence, and co-founded the JiZhi Academic Community “Urban Science Reading Club.” Recipient of the ACM SIGSPATIAL China Chapter Outstanding Paper nomination award. Authored one academic monograph, published 18 scholarly papers, with representative works appearing in top CCF A‑rank AI and data mining venues such as IEEE TPAMI, IEEE TKDE, ACM SIGKDD, AAAI, etc., accumulating over 1,100 citations.

References

1. JI J, WANG J, MOU Y, et al. How to break it down for building it up? Theory-guided graph decomposition learning for spatiotemporal traffic prediction[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2026, 48(5): 5442-5459.

2. JI J, ZHANG W, WANG J, et al. Seeing the unseen: Learning basis confounder representations for robust traffic prediction[C]//Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1. 2025: 577-588.

3. WANG J, JI J, JIANG Z, et al. Traffic flow prediction based on spatiotemporal potential energy fields[J]. IEEE Transactions on Knowledge and Data Engineering, 2023, 35(9): 9073-9087.

4. JI J, WANG J, JIANG Z, et al. STDEN: Towards physics-guided neural networks for traffic flow prediction[C]//Proceedings of the AAAI Conference on Artificial Intelligence. 2022, 36(4): 4048-4056.

Event Preview

⏰ Time: Thursday, November 5, 2026, 19:30-21:30 

📍 Online: Tencent Meeting 

📍 Live: Simultaneous broadcast on WeChat Channels and Jizhi Club Bilibili channel

If you’re interested, please scan the QR code to sign up for the reading club. After registering, you can join the participants’ group for discussion.

Live Broadcast Information

Sign up for the reading group: “Computational Demography”

To understand the complex system of population—how countless individual choices give rise to macro-level patterns and how they influence public policy,Sanhe Li Tong Technology and Jizhi Club jointly launched the Computational Demography Reading Group, with Professors Li Handong and Cui Xuefeng from the School of Systems Science at Beijing Normal University serving as co‑initiators.

The reading group will start on September 17, 2026, meeting online every Thursday from 19:30 to 21:30, expected to run for 8–10 weeks. It will include keynote presentations, discussions, and provide video recordings after each session. We warmly invite researchers in related fields and anyone with interdisciplinary interest to join.

Scan the QR code in the poster to sign up for the book club

Registration Method

Step 1: Scan the QR code with WeChat to fill in the registration information.

Scan QR code to register

Step 2: Fill in the information and register.

Step 3: Add the operations manager’s WeChat and invite them to the relevant themed reading group community (WeChat group).

PS:

We encourage in‑depth discussions around complex systems, computational demography and related specific issues. To ensure discussion quality, please avoid posting content that is unrelated to this session’s reading group theme, lacks empirical basis, or is overly abstract philosophical speculation.

If the discussion clearly deviates from the topic and does not adjust after a reminder from the host, we will have to remove that member from the discussion group to maintain the overall learning environment, and refund the unused portion proportionally based on their actual participation.

Thank you for your understanding and cooperation. Let’s create a focused, deep, and rewarding co‑learning space together.

Join the community to enjoy core resources

Become a member to unlock the full learning ecosystem, including:online real-time Q&A, full course recordings for replay, exclusive material sharing, high-quality community interaction, first-hand information sync, and earn points by participating in co-creation tasks among other benefits.

Special Refund and Incentive Mechanism

We offer the following two ways for your investment to receive tangible returns:

1. Task Completion Refund Path
   
   Claim and successfully complete any two subtitle tasks, and you will receive a full refund of the registration fee, plus an exclusive JiZhi merchandise reward.
2. Operational Growth Incentive Path
   
   After successfully completing one subtitle task, you can apply to become an operations assistant. Upon the successful completion of the reading club project, the tuition will be refunded. Outstanding performers may also have the chance to receive an additional scholarship.

Recommended Reading

City Science Book Club Launch: AI and Causality Intertwined Urban Complex Systems

The Law of City Scale: Allometric Growth in Urban Studies and Spatial Science | New Book Brief

Cities as Complex Living Systems: Complexity Theory Guides Urban Digital Twin Models

Nature: Human Mobility Networks Reveal Growing Segregation in Large Cities

How to Predict Traffic Congestion? A Travel Guide from Complex Systems

The Era of High-Quality Population Development Has Arrived: Understand the New “15th Five-Year” Plan for Fertility, Education, Health, and Elderly Care in One ArticleDeployment

Sichuan’s aging population is accelerating; how can social elderly care services meet the retirement needs of 21.5 million people?

Kickstarting the 14th Five-Year Plan: How household registration reform promotes equalization of public services—let “numbers follow people” precede “money follow people”

Click “Read Original” to sign up for the book club