Neil He

Neil He

he/him

PhD Student in Computer Science at UIUC

Posts by Collection

portfolio

publications

Position: Beyond Euclidean – Foundation Models Should Embrace Non-Euclidean Geometries

Published in preprint, under review, 2025

This paper argues for the necessity of incoporating non-Euclidean geometry into foundation models design, with arguments grounded in both theoretics and pratical considerations

Recommended citation: Neil He, Jiahong Liu, Buze Zhang, Ngoc Bui, Ali Maatouk, Menglin Yang, Irwin King, Melanie Weber, and Rex Ying. "Position: Beyond Euclidean -- Foundation Models Should Embrace Non-Euclidean Geometries." arXiv preprint. 2025.
Paper Link

HELM: Hyperbolic Large Language Models via Mixture-of-Curvature Experts

Published in preprint, under review, 2025

This paper propose a framework for a family of hyperbolic LLMs, including a mixture-of-curvature experts module where each expert operates in a distinct curvature space, hyperbolic Multi-Head Latent Attention mechanism, and hyperbolic rotary positional encoding.

Recommended citation: Neil He, Rishabh Anand, Hiren Madhu, Ali Maatouk, Smita Krishnaswamy, Leandros Tassiulas, Menglin Yang, and Rex Ying. "HELM: Hyperbolic Large Language Models via Mixture-of-Curvature Experts." arXiv preprint. 2025.
Paper Link | Code Link

HyperCore: The Core Framework for Building Hyperbolic Foundation Models with Comprehensive Modules

Published in TheWebConf NEGEL Workshop, 2025

HyperCore is an easy-to-use, open source library for building hyperbolic deep learning networks, especially hyperbolic foundation models. This paper details the library components and builds the several novel hyperbolic foundation models with it, such as hyperbolic CLIP, ViT, and GraphRAG.

Recommended citation: Neil He, Menglin Yang, and Rex Ying. "HyperCore: The Core Framework for Building Hyperbolic Foundation Models with Comprehensive Modules." TheWebConf NEGEL Workshop. 2025.
Paper Link | Code Link

Efficient Diffusion Models for Symmetric Manifolds

Published in The Forty-Second International Conference on Machine Learning (ICML), 2025

This paper is proposes an efficient diffusion framework for generative modelling on symmetric manifolds, such as the unitary group, with provable convergence guarantees and iteration complexity.

Recommended citation: Oren Mangoubi, Neil He, and Nisheeth K. Vishnoi. "Efficient Diffusion Models for Symmetric Manifolds." in ICML. 2025
Paper Link | Code Link

Hyperbolic Deep Learning for Foundation Models: A Survey

Published in ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2025

This paper presents a comprehensive survey that details the recent techinical advanvents in hyperbolic foundation models.

Recommended citation: Neil He, Hiren Madhu, Ngoc Bui, Menglin Yang, and Rex Ying. "Hyperbolic Deep Learning for Foundation Models: A Survey." in KDD. 2025.
Paper Link | Download Slides

Lorentzian Residual Neural Networks

Published in ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2025

This paper proposes an efficient, stable, and effective residual neural network framework in hyperbolic space.

Recommended citation: Neil He, Menglin Yang, and Rex Ying. 2025. Lorentzian Residual Neural Networks. In KDD .
Paper Link | Code Link

talks

Hyperbolic Deep Learning for Foundation Models: A Tutorial

Published:

More information here I’m leading a tutorial for hyperbolic foundation models at SIGKDD 2025! This tutorial aims to provide a comprehensive understanding of hyperbolic deep learning methods espeically their application to foundation models. It will cover the theoretical foundations, practical implementations, and future research directions in this exciting field. The tutorial is designed for a broad audience, including both newcomers and experts in machine learning, and will feature interactive components to engage participants.

Non-Euclidean Foundation Models: Advancing AI Beyond Euclidean Frameworks

Published:

More information here I’m orgranizing a workshop for non-Euclidean Foundation Models at NeurIPS 2025! This workshop will include submissions, talks, and poster sessions on topics related to the intersection of foundation models and non-Euclidean representation learning, including theoretical foundations, architectures and algorithms, applications, trustworthiness and robustness, and benchmarks and tools.

teaching