KDD 2026 Workshop: Geometric Space, Architecture and Learning Objectives for Large Pre-Trained Models
The Geometric Space, Architecture and Learning Objectives for Large Pre-Trained Models (GALOP) workshop at KDD 2026 brings together researchers working on geometric representation spaces, geometry-aware architectures, and learning objectives for large pre-trained models, spanning natural language processing, computer vision, graph learning, knowledge discovery, and scientific AI.
News
- 2026-08-09: 🔥 The final workshop program and Workshop Slack group are now available.
- 2026-07-25: Workshop date and time information has been updated.
- 2026-06-10: Paper notifications have been sent through OpenReview. Camera-ready revision is open for accepted papers.
- 2026-03-22: Call for papers, OpenReview portal
- 2026-03-11: The workshop proposal was accepted by KDD 2026 as a half day workshop.
Introduction
Foundation models now drive progress across language, vision, graphs, recommendation, and scientific discovery, yet most of them are still built around Euclidean representations and objectives. In many real world settings, however, the underlying data contain hierarchy, relational structure, multi scale organization, or nonuniform geometry that is not naturally captured by standard design choices.
Target. This workshop focuses on how geometric spaces, geometric neural networks, and geometric objectives can improve foundation models by introducing more appropriate inductive bias for representation, reasoning, and adaptation. It highlights work on hyperbolic, spherical, mixed curvature, and other geometry aware approaches that can better align model structure with the structure of data.
Goal. The goal is to bring together researchers from machine learning, data mining, natural language processing, computer vision, graph learning, knowledge discovery, and scientific AI. By connecting theory, methods, systems, and applications, the workshop aims to create a shared forum for understanding when geometric modeling matters, how it should be integrated into large models, and how its benefits should be evaluated in practice.
Important Dates
Time: 11:59 PM Anywhere on Earth unless otherwise specified.
- Workshop paper submission:
April 30, 2026extended to May 31, 2026 - Workshop paper notification: June 10, 2026
- Camera-ready revision: June 15, 2026
- Final workshop program, materials, and full website online: June 22, 2026
- Workshop date: August 9, 2026 afternoon (1:00 pm-5:00 pm)
- Conference dates: August 9 to 13, 2026
- Venue: Room 201B, International Convention Center Jeju (ICC Jeju)
Topics of Interest
We welcome submissions on topics including, but not limited to, the following directions from the proposal:
- Hyperbolic, spherical, and mixed curvature embeddings for large pre-trained models
- Non Euclidean word, sentence, document, and multimodal representations
- Geometric transformers and manifold aware attention mechanisms
- Equivariant and invariant architectures for foundation models
- Metric learning and contrastive objectives with geometric constraints
- Curvature aware optimization on Riemannian manifolds
- Alignment and fusion across different geometric spaces
- Theory for geometric large pre-trained models, including expressiveness and generalization
- Applications in natural language processing, computer vision, graph learning, knowledge discovery, and scientific discovery
- Benchmarks, evaluation protocols, open source tools, visualization, and reproducibility resources
Submission (Completed)
- The submission period has closed.
- Authors of accepted papers should submit the camera-ready revision through OpenReview by June 15, 2026.
- Accepted papers will be presented as oral or poster presentations, as indicated in the decision notification.
- The workshop follows the current KDD 2026 workshop policy and is planned as an in person event.
GALOP Workshop @ KDD 2026 OpenReview Portal
Accepted Papers
- Geometric Perturbation Graph Neural Network for Heterophily Modeling (Oral + Poster)
- Hyperattentive Residuals (Oral + Poster)
- Revisiting CF-Integrated LLM Recommenders through Dimensional Collapse (Oral + Poster)
- MeSH-HyRerank: Hyperbolic Ontology Adaptation for Biomedical Foundation Model Retrieval (Oral + Poster)
- Curvature-Adaptive Self-Attention: Riemannian Transformers with Distortion and Generalization Bounds (Poster)
- Extracting Local Manifold Geometry from Pretrained Diffusion Models in One Inverse Step (Poster)
Program
| Time (KST) | Session | Speaker / Presentation |
|---|---|---|
| 1:00–1:05 PM | Opening Remarks | — |
| 1:05–2:00 PM | Invited Talk 1 (Geometric Objective Function) |
Prof. Yifei Zhang Understanding Self-supervised Learning from the Dimensional Collapse Perspective |
| 2:00–3:00 PM | Invited Talk 2 (Geometric Architecture) |
Prof. Kijung Shin AI for Complex Networks: With a Focus on Hypergraphs |
| 3:00–3:30 PM | Coffee Break and Poster Session | — |
| 3:30–3:45 PM | Contributed Talk 1 | Geometric Perturbation Graph Neural Network for Heterophily Modeling |
| 3:45–4:00 PM | Contributed Talk 2 | MeSH-HyRerank: Hyperbolic Ontology Adaptation for Biomedical Foundation Model Retrieval |
| 4:00–4:15 PM | Contributed Talk 3 | Revisiting CF-Integrated LLM Recommenders through Dimensional Collapse |
| 4:15–4:30 PM | Contributed Talk 4 | Hyperattentive Residuals |
| 4:30–5:00 PM | Organizational Talk (Geometric Embedding Space) |
Ms. Jiahong Liu Hyperbolic Learning in the Era of Large Language Models |
Invited Speakers
Yifei Zhang
Professor, School of Computer Science, Northwestern Polytechnical University
Yifei Zhang works on trustworthy machine learning, federated learning, graph representation learning, large language models, and robust vision-language models.
Kijung Shin
Associate Professor, KAIST AI & EE; Director, Data Mining Lab
Kijung Shin works on data mining, graph algorithms, and network science, with broad interests in scalable methods for structured and relational data.
Organizers
Menglin Yang HKUST (GZ) |
Jiahong Liu CUHK |
Lucas Vinh Tran JPMorgan Chase |
Rex Ying Yale University |
Slack Group
Join the Workshop Slack group for community discussions and updates.
Contact
For questions about the workshop, please contact us at galop-kdd2026@googlegroups.com.