mlsanityData to 14 Oct 2026

PANDA: Prior-guided Attentional Dual-path Architecture

arXiv 2610.13635 ↗First posted 9 Oct 2026cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

Hypergraph Neural Networks (HNNs), designed to model higher-order relations beyond pairwise interactions, typically rely either on fixed structural weights or on normalised attention. Fixed weights can propagate misleading information in heterophilic settings, while normalisation makes attention coefficients competitive, in the sense that, in a message-passing framework, increasing the weight of one sender must reduce the weight of others in the same aggregation set. Our main contribution is the introduction of PANDA (Prior-guided AttentioNal Dual-path Architecture), a plug-in for two-stage message passing HNNs applicable to both undirected and directed settings. The main path interpolates between learned receiver-normalised attention and structural incidence priors, while a non-competitive (not normalised) auxiliary path lets each sender contribute its own transformed information. We also show that, with fixed structural coefficients, PANDA recovers a certain complex-valued convolution operator adopted in previous works. Across four HNNs spanning spatial and spectral methods in both directed and undirected settings, plugging PANDA in yields an average improvement of 5.18 percentage points across eight real-world benchmark datasets. The code is available at https://github.com/jainris/panda.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 32 of its 50 references that are in the map. First posted on arXiv 9 Oct 2026.

Builds on

32 of 50 references.

References · 32

  1. 2017Graph Attention Networkscited by 8,923
  2. 2020GLU Variants Improve Transformercited by 1,804
  3. 2021How Attentive are Graph Attention Networks?cited by 835
  4. 2016Language Modeling with Gated Convolutional Networkscited by 702
  5. 2023Gated Linear Attention Transformers with Hardware-Efficient Trainingcited by 514
  6. 2018Hypergraph Neural Networkscited by 488
  7. 2023A critical look at the evaluation of GNNs under heterophily: Are we really making progress?cited by 343
  8. 2025Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Freecited by 283
  9. 2019Stabilizing Transformers for Reinforcement Learningcited by 270
  10. 2015Highway Networkscited by 260
  11. 2019Hypergraph Convolution and Hypergraph Attentioncited by 228
  12. 2022Transformer Quality in Linear Timecited by 191
20 more
  1. 2018GaAN: Gated Attention Networks for Learning on Large and Spatiotemporal Graphscited by 165
  2. 2021UniGNN: a Unified Framework for Graph and Hypergraph Neural Networkscited by 130
  3. 2021You are AllSet: A Multiset Function Framework for Hypergraph Neural Networkscited by 127
  4. 2016Higher-order organization of complex networkscited by 110
  5. 2020HNHN: Hypergraph Networks with Hyperedge Neuronscited by 109
  6. 2018Simplicial Closure and higher-order link predictioncited by 93
  7. 2020Be More with Less: Hypergraph Attention Networks for Inductive Text Classificationcited by 74
  8. 2022Equivariant Hypergraph Diffusion Neural Operatorscited by 47
  9. 2020Bridging the Gap Between Spectral and Spatial Domains in Graph Neural Networkscited by 27
  10. 2022SigMaNet: One Laplacian to Rule Them Allcited by 24
  11. 2020Semi-supervised Hypergraph Node Classification on Hypergraph Line Expansioncited by 22
  12. 2020Normalized Attention Without Probability Cagecited by 20
  13. 2020Nonlinear Higher-Order Label Spreadingcited by 20
  14. 2022Preventing Over-Smoothing for Hypergraph Neural Networkscited by 20
  15. 2021Bridging the Gap between Spatial and Spectral Domains: A Unified Framework for Graph Neural Networkscited by 19
  16. 2023From Hypergraph Energy Functions to Hypergraph Neural Networkscited by 18
  17. 2022Hypergraph Convolutional Networks via Equivalency between Hypergraphs and Undirected Graphscited by 16
  18. 2023Hypergraph Neural Networks through the Lens of Message Passing: A Common Perspective to Homophily and Architecture Designcited by 15
  19. 2025Sigmoid Self-Attention has Lower Sample Complexity than Softmax Self-Attention: A Mixture-of-Experts Perspectivecited by 4
  20. 2026Pairwise is Not Enough: Hypergraph Neural Networks for Multi-Agent Pathfindingcited by 4

Built on by

Not cited yet.