Pith. sign in

REVIEW 1 cited by

Origin-Destination Network Generation via Gravity-Guided GAN

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2306.03390 v1 pith:2AWNUH4Q submitted 2023-06-06 cs.LG

Origin-Destination Network Generation via Gravity-Guided GAN

classification cs.LG
keywords urbanmethodsorigin-destinationphysicsbuildconsidereveryfeatures
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Origin-destination (OD) flow, which contains valuable population mobility information including direction and volume, is critical in many urban applications, such as urban planning, transportation management, etc. However, OD data is not always easy to access due to high costs or privacy concerns. Therefore, we must consider generating OD through mathematical models. Existing works utilize physics laws or machine learning (ML) models to build the association between urban structures and OD flows while these two kinds of methods suffer from the limitation of over-simplicity and poor generalization ability, respectively. In this paper, we propose to adopt physics-informed ML paradigm, which couple the physics scientific knowledge and data-driven ML methods, to construct a model named Origin-Destination Generation Networks (ODGN) for better population mobility modeling by leveraging the complementary strengths of combining physics and ML methods. Specifically, we first build a Multi-view Graph Attention Networks (MGAT) to capture the urban features of every region and then use a gravity-guided predictor to obtain OD flow between every two regions. Furthermore, we use a conditional GAN training strategy and design a sequence-based discriminator to consider the overall topological features of OD as a network. Extensive experiments on real-world datasets have been done to demonstrate the superiority of our proposed method compared with baselines.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. OpFlow: Learning Opportunity-Conditioned Choice Potentials for Robust OD Flow Prediction

    cs.LG 2026-07 conditional novelty 6.5

    Robust OD flow prediction comes from learning row-centered exposure-to-choice potentials and reconstructing counts as scale times allocation, not from raw-count supervision.