Intrinsic Lorentz Neural Network

• URL: http://arxiv.org/abs/2602.23981v1
• Date: Fri, 27 Feb 2026 12:48:05 GMT
• Title: Intrinsic Lorentz Neural Network
• Authors: Xianglong Shi, Ziheng Chen, Yunhan Jiang, Nicu Sebe,
• Abstract summary: Real-world data frequently exhibit latent hierarchical structures, which can be naturally represented by hyperbolic geometry.<n>We propose the emphIntrinsic Lorentz Neural Network (ILNN), a fully intrinsic hyperbolic architecture that conducts all computations within the Lorentz model.<n>At its core, the network introduces a novel emphpoint-to-hyperplane fully connected layer (FC), replacing traditional Euclidean affine logits with closed-form hyperbolic distances.
• Score: 49.83037691286893
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Real-world data frequently exhibit latent hierarchical structures, which can be naturally represented by hyperbolic geometry. Although recent hyperbolic neural networks have demonstrated promising results, many existing architectures remain partially intrinsic, mixing Euclidean operations with hyperbolic ones or relying on extrinsic parameterizations. To address it, we propose the \emph{Intrinsic Lorentz Neural Network} (ILNN), a fully intrinsic hyperbolic architecture that conducts all computations within the Lorentz model. At its core, the network introduces a novel \emph{point-to-hyperplane} fully connected layer (FC), replacing traditional Euclidean affine logits with closed-form hyperbolic distances from features to learned Lorentz hyperplanes, thereby ensuring that the resulting geometric decision functions respect the inherent curvature. Around this fundamental layer, we design intrinsic modules: GyroLBN, a Lorentz batch normalization that couples gyro-centering with gyro-scaling, consistently outperforming both LBN and GyroBN while reducing training time. We additionally proposed a gyro-additive bias for the FC output, a Lorentz patch-concatenation operator that aligns the expected log-radius across feature blocks via a digamma-based scale, and a Lorentz dropout layer. Extensive experiments conducted on CIFAR-10/100 and two genomic benchmarks (TEB and GUE) illustrate that ILNN achieves state-of-the-art performance and computational cost among hyperbolic models and consistently surpasses strong Euclidean baselines. The code is available at \href{https://github.com/Longchentong/ILNN}{\textcolor{magenta}{this url}}.

Related papers

• Hyperbolic Busemann Neural Networks [100.27390895360753]
  In this work, we lift two core components of neural networks, Multinomial Logistic Regression (MLR) and Fully Connected (FC) layers, into hyperbolic space via Busemann functions.<n> BMLR provides compact parameters, a point-to-horosphere distance interpretation, batch-efficient computation, and a Euclidean limit.<n>Experiments on image classification, genome sequence learning, node classification, and link prediction demonstrate improvements in effectiveness and efficiency over prior hyperbolic layers.
  arXiv  Detail & Related papers  (2026-02-21T14:45:40Z)
• Fast and Geometrically Grounded Lorentz Neural Networks [44.564864487582525]
  We develop a formulation of hyperbolic neural networks that is both efficient and captures the key properties of hyperbolic space.<n>We prove that, with the current formulation of Lorentz linear layers, the hyperbolic norms of the outputs scale logarithmically with the number of gradient descent steps.<n>Our new formulation, together with further efficiencies through Lorentzian activation functions and a new caching strategy results in neural networks fully abiding by hyperbolic geometry.
  arXiv  Detail & Related papers  (2026-01-29T10:44:32Z)
• Lorentz Local Canonicalization: How to Make Any Network Lorentz-Equivariant [9.946772087556333]
  Lorentz-equivariant neural networks are becoming the leading architectures for high-energy physics.<n>We introduce Lorentz Local Canonicalization (LLoCa), a general framework that renders any backbone network exactly Lorentz-equivariant.<n>Our models achieve competitive and state-of-the-art accuracy on relevant particle physics tasks, while being $4times$ faster and using $10times$ fewer FLOPs.
  arXiv  Detail & Related papers  (2025-05-26T17:57:17Z)
• Can we ease the Injectivity Bottleneck on Lorentzian Manifolds for Graph Neural Networks? [0.0]
  Lorentzian Graph Isomorphic Network (LGIN) is a novel HGNN designed for enhanced discrimination within the Lorentzian model.<n>LGIN is the first to adapt principles of powerful, highly discriminative GNN architectures to a Riemannian manifold.
  arXiv  Detail & Related papers  (2025-03-31T18:49:34Z)
• Hyperbolic Binary Neural Network [17.84584119915958]
  Binary Neural Network (BNN) converts full-precision weights and activations into their extreme 1-bit counterparts.<n>This paper introduces the Hyperbolic Binary Neural Network (HBNN) by leveraging the framework of hyperbolic geometry to optimize the constrained problem.
  arXiv  Detail & Related papers  (2025-01-07T02:15:58Z)
• Kronecker-Factored Approximate Curvature for Modern Neural Network Architectures [85.76673783330334]
  Two different settings of linear weight-sharing layers motivate two flavours of Kronecker-Factored Approximate Curvature (K-FAC) We show they are exact for deep linear networks with weight-sharing in their respective setting. We observe little difference between these two K-FAC variations when using them to train both a graph neural network and a vision transformer.
  arXiv  Detail & Related papers  (2023-11-01T16:37:00Z)
• A Unification Framework for Euclidean and Hyperbolic Graph Neural Networks [8.080621697426997]
  Hyperbolic neural networks can effectively capture the inherent hierarchy of graph datasets. They entangle multiple incongruent (gyro-)vector spaces within a layer, which makes them limited in terms of generalization and scalability. We propose the Poincare disk model as our search space, and apply all approximations on the disk. We demonstrate that our model not only leverages the power of Euclidean networks such as interpretability and efficient execution of various model components, but also outperforms both Euclidean and hyperbolic counterparts on various benchmarks.
  arXiv  Detail & Related papers  (2022-06-09T05:33:02Z)
• Fully Hyperbolic Neural Networks [63.22521652077353]
  We propose a fully hyperbolic framework to build hyperbolic networks based on the Lorentz model. We show that our method has better performance for building both shallow and deep networks.
  arXiv  Detail & Related papers  (2021-05-31T03:36:49Z)
• Hyperbolic Variational Graph Neural Network for Modeling Dynamic Graphs [77.33781731432163]
  We learn dynamic graph representation in hyperbolic space, for the first time, which aims to infer node representations. We present a novel Hyperbolic Variational Graph Network, referred to as HVGNN. In particular, to model the dynamics, we introduce a Temporal GNN (TGNN) based on a theoretically grounded time encoding approach.
  arXiv  Detail & Related papers  (2021-04-06T01:44:15Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.