Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

Sheaves Reloaded: A Directional Awakening

T0 review · 4 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read The paper claims that a Directed Sheaf Laplacian carries edge direction through a complex phase and that DSNN, the first sheaf network with this directional bias, consistently outperforms GNN and SNN baselines on nine node-classification…

desk verdict A genuinely useful directed sheaf architecture with a sound core, though the theorem statements overstate the match to the Magnetic Laplacian and the empirical claims need calibration. read the letter →

arxiv 2506.02842 v1 pith:WT5NZZGG submitted 2025-06-03 cs.LG cs.AI

Stefano Fiorini , Hakan Aktas , Iulia Duta , Stefano Coniglio , Pietro Morerio , Alessio Del Bue , Pietro Liò This is my paper
classification cs.LGcs.AI MSC 05C5055N3068T07
keywords directedcellularsheafLaplacianneuralnetworksgraphedgedirectionalityheterophilicgraphsmagneticsign-magnetic
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Sheaf neural networks move features between nodes through linear restriction maps, but until now those maps have ignored which way an edge points. This paper introduces the Directed Cellular Sheaf, in which the restriction map at the tail of a directed edge is multiplied by the unit-modulus phase $T^{(q)}=\exp(i2\pi q(A-A^\top))$, so the edge's direction is written into the algebra of the sheaf rather than into a separate aggregation rule. From this sheaf the authors define the Directed Sheaf Laplacian $L^{\tilde F}$ and the Directed Sheaf Neural Network (DSNN), which they describe as the first sheaf architecture with an explicit directional inductive bias. Their stated result is that DSNN consistently outperforms a broad set of GNN and SNN baselines on nine real-world node-classification benchmarks, with the largest gains on heterophilic graphs, and that on synthetic directed stochastic block models the non-directional sheaf baselines collapse while DSNN retains high accuracy. A sympathetic reader should care because the construction shows directionality can be added to sheaf models with a single algebraic change, at a constant computational overhead, while recovering existing magnetic-Laplacian models as special cases.

What carries the argument

The load-bearing object is the Directed Sheaf Laplacian $L^{\tilde F}=\tilde\delta^*\tilde\delta$, built from a Directed Cellular Sheaf whose edge restriction maps are $F_{u\unlhd e}\in\mathbb{R}^{d\times d}$ at the head and $F_{v\unlhd e}=F_{v\unlhd e}^{0}\,T^{(q)}_{uv}$ at the tail. The directional memory is the Hermitian matrix $T^{(q)}=\exp(i2\pi q(A-A^\top))$, which equals $1$ on undirected edges and $-i$ on directed edges at $q=1/4$, making the off-diagonal blocks of the Laplacian complex and orientation-sensitive while the spectrum remains real and nonnegative. DSNN's update is the discretized diffusion $X^{(t+1)}=\mathrm{diag}(1+\varepsilon)X^{(t)} - \sigma(\tilde L^{\tilde F}_N (I_n\otimes W_1)X^{(t)}W_2)$, with restriction maps learned from concatenated node features and a complex ReLU activation. The theoretical bridge to existing models is Theorem 4: a trivial directed sheaf makes $L^{\tilde F}$ equal to the Magnetic Laplacian, and at $q=1/4$ to the Sign-Magnetic Laplacian.

What would settle it

Take any directed graph, randomly permute the orientations of its edges so $A-A^\top$ changes while the undirected skeleton and degrees stay fixed, and compare DSNN's accuracy with $q=1/4$ against the same model with $q=0$: the directional-bias claim predicts a clear accuracy drop, and if accuracy is unchanged, the phase encoding is not carrying the directional signal.

Watch

Extended reading notes

Core claim

The paper's central claim is that direction is not an extra ingredient to bolt onto a sheaf; it can be placed inside the restriction maps themselves. A Directed Cellular Sheaf assigns complex vector spaces to nodes and edges, keeps the head node's restriction map real, and sets the tail node's map to a real map followed by the Hermitian phase $T^{(q)}=\exp(i2\pi q(A-A^\top))$. At $q=1/4$, a directed edge $(u,v)$ contributes a factor $-i$, an undirected edge contributes $1$, and the resulting Directed Sheaf Laplacian $L^{\tilde F}=\tilde\delta^*\tilde\delta$ is Hermitian positive semidefinite with normalized spectrum in $[0,2]$. The authors prove that for a trivial one-dimensional sheaf this operator coincides with the Magnetic Laplacian $L^{(q)}$ and, at $q=1/4$, with the Sign-Magnetic Laplacian, and they give both of those matrices a new node-to-edge incidence factorization $L^{(q)}=\hat B\hat B^*$. On top of this operator, DSNN learns its restriction maps from node features and diffuses the node signal through the normalized Directed Sheaf Laplacian, which is the mechanism behind the reported gains on the nine benchmarks.

Load-bearing premise

The load-bearing premise is that a directed edge's meaning is fully captured by multiplying the tail node's restriction map by the complex phase $e^{i2\pi q(A-A^\top)}$, so any directional signal that only a non-Hermitian, non-normal operator could see is invisible to this Laplacian.

Editorial extensions

If this is right

  • DSNN reduces to NSD on undirected graphs and to MagNet and SigMaNet in the trivial-sheaf limit, so those architectures become special cases of one operator.
  • The incidence factorization $L^{(q)}=\hat B\hat B^*$ gives the Magnetic and Sign-Magnetic Laplacians a Hermitian coboundary decomposition, which yields simpler positive-semidefiniteness proofs than the original constructions.
  • On the nine real benchmarks, DSNN is reported to beat the second-best baseline on seven of nine and every directional GNN baseline on all nine, with the largest single gain over directional baselines of 15.59 points on Cornell.
  • On synthetic DSBM graphs, DSNN variants hold 86-96% accuracy while all non-directional sheaf baselines except one fall to chance-level 20%, supporting the claim that the directional bias, not complex parameters alone, drives the improvement.
  • Adding the directional phase costs only a constant multiplicative factor (about 4) in the forward pass, leaving DSNN's asymptotic complexity equal to NSD's.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Because the phase encoding lives in the sheaf rather than in a model-specific layer, the same trick could be dropped into any spectral pipeline that currently symmetrizes a directed graph: replace the symmetrized adjacency with $A_s\odot T^{(q)}$ and inherit a tunable directional bias.
  • Theorem 4's factor-of-two gap between $L^{\tilde F}$ and $L^{(q)}$ on digon-free graphs suggests the choice of degree normalization is not neutral; ablating in-degree vs out-degree vs symmetrized degree could separate true directional signal from degree asymmetry.
  • A control the paper does not run would be to randomly permute edge orientations while keeping the graph skeleton fixed; DSNN's stated mechanism predicts accuracy should drop, and measuring that drop would isolate the directional component.
  • The construction is stated for complex vector spaces and ordinary graphs, but the same Hermitian phase can be attached to restriction maps on hyperedges or higher-dimensional cells, a direction the authors list as future work but do not test.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper introduces Directed Cellular Sheaves, in which edge orientation is encoded into complex-valued restriction maps through a Hermitian phase matrix T^(q)=exp(i2πq(A-A^T)), and defines the Directed Sheaf Laplacian L^F=δ*δ together with its normalized version. On this operator the authors build the Directed Sheaf Neural Network (DSNN), a sheaf-diffusion architecture with a directional inductive bias, and report node-classification experiments on nine real-world benchmark datasets plus a synthetic DSBM study. The paper's central claims are that L^F captures both topology and direction, that it generalizes the Magnetic and Sign-Magnetic Laplacians, and that DSNN consistently outperforms baseline GNNs and SNNs.

Significance. The construction is attractive and largely sound: L^F is Hermitian and PSD by construction, its normalized version has spectrum in [0,2], and the architecture follows naturally from the directed sheaf diffusion equation. The synthetic DSBM experiment cleanly separates DSNN from undirected sheaf baselines, and the real-world results are strong enough to be of interest to the graph-learning community. If the claims are corrected, the paper would provide a principled way to inject directional information into sheaf neural networks and a useful incidence-matrix decomposition of magnetic-style Laplacians. However, the exact-equality theorems are overstated as written, and the empirical isolation of the directional mechanism is incomplete, so the advertised 'directional awakening' is currently stronger than what the manuscript demonstrates.

major comments (4)
  1. [Section 3.3 (Theorem 4)] Theorem 4 states that L^F 'coincides' with the Magnetic Laplacian L^(q), but the proof in Part 2b derives L^F = 2 L^(q) for directed graphs without digons, and a similar factor of 2 arises when digons are present. The theorem as stated is therefore false; the correct unnormalized statement is L^F = 2 L^(q), while the normalized operators coincide for digon-free graphs. Since the DSNN uses the normalized Laplacian, the architectural relationship is salvageable, but every statement in the paper that relies on exact equality, including the abstract's and conclusion's claims that DSNN generalizes MagNet and SigMaNet, must be rephrased as equality up to the scaling factor.
  2. [Section 3.3 (Theorem 5)] Theorem 5 asserts that L^(q) = B B* for a trivial directed cellular sheaf. The proof, however, first establishes L^F = B B* and then invokes Theorem 4; because Theorem 4's exact equality is false, the claimed incidence-matrix decomposition of L^(q) holds only up to the same factor of 2 (or, equivalently, for the normalized version). The theorem should be restated and re-proved with the scaling made explicit, since this decomposition is one of the paper's advertised theoretical contributions.
  3. [Definition 1 and Eq. (3)] The directional information carried by L^F is confined to the off-diagonal phase T^(q): the degree diagonal D_uu is a sum over all incident edges of F*_{u⊴e}F_{u⊴e}, so in-degree and out-degree contribute identically, and reversing every edge orientation conjugates L^F and leaves its spectrum unchanged. The manuscript's phrasing that L^F 'faithfully integrates the sign and directionality' (Section 1) and 'captures both topology and directional information' is therefore stronger than what the operator actually represents. This limitation should be stated explicitly, and the claim tempered unless a control experiment demonstrates that the phase information, rather than the learned sheaf parameterization, drives the empirical gains.
  4. [Section 5.2 (Figure 2)] The synthetic DSBM experiment compares DSNN only against NSD and NSD-Compl, which either ignore direction or use complex restriction maps without a magnetic phase. It does not include MagNet, SigMaNet, or DirGNN, the baselines that already carry a directional inductive bias. As a result, the conclusion that the large observed gains 'underscore DSNN's ability to leverage edge directionality' is not fully isolated: a simpler explanation is that any complex/magnetic encoding helps on this synthetic task, or that the sheaf parameterization is independently beneficial. Adding at least one directed spectral baseline (or a reversed-orientation control) would make the directional-bias claim supported by the controlled experiment.
minor comments (5)
  1. [Section 2.1] The sentence 'mirroring a similar property that haolds for the standard graph Laplacian' contains a typo: 'haolds' should be 'holds'.
  2. [Section 3.1 (Definition 1 rationale)] At q=1/4, the definition T^(q)_{uv}=exp(i2πq(Auv-Avu)) with Auv=1 and Avu=0 gives T^(q)_{uv}=exp(iπ/2)=+i, but the text writes cos(-π/2)+i sin(-π/2)=-i. The indexing convention should be made consistent in this explanatory passage.
  3. [Theorem 4 proof (Parts 2a and 2b)] The proof states twice that 'the last equation holds since T^(q)_{uu}=0 for any q'. In fact T^(q)_{uu}=exp(0)=1; the term vanishes because A_s,uu=0, not because T^(q)_{uu}=0. This typographical error should be corrected.
  4. [Section 3.1, text after Eq. (2)] The phrase 'for a directed edge e=(u,v) or e=(v,e)' contains a typo: the second case should be e=(v,u), not e=(v,e).
  5. [Appendix A] The paper promises that datasets and code are available on GitHub, but Appendix A does not actually provide the repository URL. Including the link would materially improve reproducibility.

Circularity Check

1 steps flagged · score 4.0 of 10

The 'generalization' to the Magnetic and Sign-Magnetic Laplacians is written into Definition 1 via the imported phase matrix T(q); the DSNN architecture and its empirical wins are independent of that definitional embedding.

  1. self definitional [Definition 1 (Section 3.1, Eqs. 2-3); Theorem 4 (Section 3.3)]
    "The Directed Cellular Sheaf ... 1. A directional and topological Hermitian matrix T^(q) := exp(i2πq(A−A^T)), with q∈R. ... eFv⊴e = eF0v⊴eT(q)uv ∈ C^{d×d}. ... Theorem 4. ... the Directed Sheaf Laplacian L eF associated with a Trivial Directed Cellular Sheaf coincides with the Magnetic Laplacian L(q). In the special case where q=1/4, L eF also coincides with the Sign-Magnetic Laplacian Lσ."

    Section 2.2 defines the Magnetic Laplacian via H^(q) := As ⊙ exp(i2πq(A−A^T)). Definition 1 imports exactly that phase as a 'directional and topological Hermitian matrix' T^(q) and inserts it into the tail node's restriction map (eFv⊴e = eF0v⊴e T^(q)uv). Equations (2)-(3) then give off-diagonal blocks −F^T_u⊴e F0_v⊴e T^(q)uv and diagonal Σ_{e∈Γ(u)} F*_u⊴e F_u⊴e, so for the trivial (d=1) sheaf the DSL becomes, entry by entry, Ds − As ⊙ T^(q) — i.e., L^(q) up to the factor 2 that Theorem 4 Part 2b proves and then declares 'immaterial'. The claimed generalization is therefore a recomputation of the defining phase choice, not a derived consequence. The genuinely new content is the learned non-trivial (d>1) sheaf case and the architecture built on it, which keeps the circularity partial.

full rationale

Most of the paper is self-contained. The Directed Sheaf Laplacian is defined as eδ*eδ; Theorem 1 (Hermitian, PSD) and Theorem 2 (⪯2I) are proven directly from that definition, and the empirical evaluation is independent: nine real-world datasets with standard external baselines (GCN, GAT, GraphSAGE, GGCN, H2GCN, GPRGNN, FAGCN, MixHop, GCNII, MagNet, SigMaNet, DirGNN, NSD) plus a controlled complex-restriction-map baseline (NSD-Compl) that the authors introduce precisely to isolate the directional bias. The one definitional shortcut is Theorem 4: the phase matrix T^(q) of Definition 1 is taken verbatim from the Magnetic Laplacian literature, so the coincidence L^F = L^(q) for a trivial sheaf is built into the definitions rather than discovered. The q=1/4 equality with the Sign-Magnetic Laplacian is inherited from the same authors' SigMaNet paper [21]; because that identity is published, externally checkable, and not the load-bearing premise of the paper's central claim, the self-citation is real evidence rather than a circular loop. For d>1 the DSL with learned restriction maps is not the Magnetic Laplacian, and the central claim that DSNN outperforms baselines rests on experiments that do not depend on Theorem 4. The critic's observation that the symmetrized-degree, real-spectrum DSL leaves the spectrum invariant under global edge reversal is a legitimate scope limitation of the directional encoding, but it concerns correctness rather than circularity. Score 4 reflects partial definitional circularity in the theory claims with an independent empirical core.

Assumptions & free parameters 3 free parameters · 4 assumptions · 2 invented entities

The central mathematical machinery relies on a small number of modeling choices: the phase encoding from the magnetic Laplacian, the finite-dimensional complex stalk assumption, and the signless relation used in the spectral bound. The only dataset-tuned scalar is q; network weights are standard learned parameters. No physical entities are introduced.

free parameters (3)
  • q (phase hyperparameter) = tuned per dataset from {0, 0.1, 0.15, 0.20, 0.25, 0.5, 0.75, 1}
    Phase parameter in T^(q) controlling the directional encoding; selected by validation accuracy, not from a theory; it determines the strength of directional bias and is dataset-specific.
  • epsilon (per-stalk magnitude) = learned in [-1,1]^d
    Eq. (8): diag(1+epsilon) rescales stalk features; a learned parameter that affects the model's dynamics.
  • Restriction-map generator weights (Phi) = end-to-end learned weights
    The restriction maps F_u⊴e = Phi(x_u || x_v) are learned network parameters (Section 4, Learnable Sheaf Laplacian); they are standard trainable weights but part of the model's capacity.
assumptions (4)
  • domain assumption The stalks are finite-dimensional complex vector spaces C^d and all restriction maps are linear.
    Assumed throughout Section 3.1; standard in the SNN literature.
  • ad hoc to paper The phase matrix T^(q)=exp(i 2 pi q (A - A^T)) with A the binary adjacency matrix defines the directional structure.
    Definition 1; this is the paper's modeling choice to encode direction through the imaginary part of the tail restriction map.
  • standard math The relation Q_N^F = 2I - L_N^F holds for the signless and signed normalized sheaf Laplacians.
    Used in the proof of Theorem 2 (Appendix C); standard for graphs, asserted for sheaves.
  • domain assumption The diagonal blocks D_uu = sum_e F_u⊴e* F_u⊴e are invertible for normalization.
    Needed in Eq. (5) and Theorems 1-2 for D^-1/2; isolated or degenerate stalks are not treated.
invented entities (2)
  • Directed Cellular Sheaf
    purpose: Mathematical structure assigning complex vector spaces and orientation-dependent complex restriction maps to nodes and edges to represent edge direction.
    Newly defined object (Definition 1); it is a construction, not a prediction; its value is judged through the resulting model's performance.
  • Directed Sheaf Laplacian L^F
    purpose: Hermitian PSD operator built as delta* delta to diffuse signals with directional bias.
    New operator defined in Section 3.1; it reduces to the magnetic Laplacian for trivial sheaves, so its directional content is not independently verified outside the model.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Sheaves Reloaded: A Directional Awakening." pith.science (2026). https://pith.science/paper/WT5NZZGG

@misc{pith2026250602842,
  author       = {Pith},
  title        = {Pith review of: Sheaves Reloaded: A Directional Awakening},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WT5NZZGG}},
  note         = {Machine review of arXiv:2506.02842}
}
read the original abstract

Sheaf Neural Networks (SNNs) represent a powerful generalization of Graph Neural Networks (GNNs) that significantly improve our ability to model complex relational data. While directionality has been shown to substantially boost performance in graph learning tasks and is key to many real-world applications, existing SNNs fall short in representing it. To address this limitation, we introduce the Directed Cellular Sheaf, a special type of cellular sheaf designed to explicitly account for edge orientation. Building on this structure, we define a new sheaf Laplacian, the Directed Sheaf Laplacian, which captures both the graph's topology and its directional information. This operator serves as the backbone of the Directed Sheaf Neural Network (DSNN), the first SNN model to embed a directional bias into its architecture. Extensive experiments on nine real-world benchmarks show that DSNN consistently outperforms baseline methods.

Figures

Figures reproduced from arXiv: 2506.02842 by the authors.

Figure 1
Figure 1. An illustration of the complex-valued restriction maps of the [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Mean and standard deviation of the accuracy on a collection of synthetic benchmarks on [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. Benchmarking Sheaf Neural Networks for Inductive Tasks

    cs.LG 2026-08 conditional novelty 6.0 of 10

    On 14 inductive graph benchmarks, sheaf neural networks underperform strong GNN baselines, and their performance is driven more by the surrounding architecture than by the sheaf diffusion mechanism.

Reference graph

Works this paper leans on

37 extracted references · 27 canonical work pages · cited by 1 Pith paper

  1. [1]

    Toward a spectral theory of cellular sheaves.Journal of Applied and Computational Topology, 3(4):315–358, 2019

    Jakob Hansen and Robert Ghrist. Toward a spectral theory of cellular sheaves.Journal of Applied and Computational Topology, 3(4):315–358, 2019

    2019

  2. [2]

    Neural sheaf diffusion: A topological perspective on heterophily and oversmoothing in gnns.Advances in Neural Information Processing Systems, 35:18527–18541, 2022

    Cristian Bodnar, Francesco Di Giovanni, Benjamin Chamberlain, Pietro Lio, and Michael Bronstein. Neural sheaf diffusion: A topological perspective on heterophily and oversmoothing in gnns.Advances in Neural Information Processing Systems, 35:18527–18541, 2022

    work page 2022

  3. [3]

    Magnet: A neural network for directed graphs, 2021

    Xitong Zhang, Yixuan He, Nathan Brugnone, Michael Perlmutter, and Matthew Hirn. Magnet: A neural network for directed graphs, 2021

    work page 2021

  4. [4]

    Local structure of directed networks

    Ginestra Bianconi, Natali Gulbahce, and Adilson E Motter. Local structure of directed networks. Physical review letters, 100(11):118701, 2008

    work page 2008

  5. [5]

    Dynamic patterns of information flow in complex networks

    Uzi Harush and Baruch Barzel. Dynamic patterns of information flow in complex networks. Nature communications, 8(1):2181, 2017

    work page 2017

  6. [6]

    Structure and dynamical behavior of non-normal networks.Science advances, 4(12):eaau9403, 2018

    Malbor Asllani, Renaud Lambiotte, and Timoteo Carletti. Structure and dynamical behavior of non-normal networks.Science advances, 4(12):eaau9403, 2018

    work page 2018

  7. [7]

    PhD thesis, Brown University, 1985

    Allen Dudley Shepard.A Cellular Description of the Derived Category of a Stratified Space. PhD thesis, Brown University, 1985

    work page 1985

  8. [8]

    PhD thesis, University of Pennsylvania, 2014

    Justin Michael Curry.Sheaves, Cosheaves and Applications. PhD thesis, University of Pennsylvania, 2014

    2014

Show all 37 references
  1. [9]

    American Mathematical Soc., 1997

    Fan RK Chung.Spectral graph theory, volume 92. American Mathematical Soc., 1997

    1997

  2. [10]

    Sheaf attention networks

    Federico Barbero, Cristian Bodnar, Haitz Sáez de Ocáriz Borde, and Pietro Lio. Sheaf attention networks. InNeurIPS 2022 Workshop on Symmetry and Geometry in Neural Representations, 2022

    2022

  3. [11]

    Sheaf hypergraph networks

    Iulia Duta, Giulia Cassarà, Fabrizio Silvestri, and Pietro Liò. Sheaf hypergraph networks. Advances in Neural Information Processing Systems, 36:12087–12099, 2023

    2023

  4. [12]

    Sheaf diffusion goes nonlinear: Enhancing gnns with adaptive sheaf laplacians.PROCEEDINGS OF MACHINE LEARNING RESEARCH, 251, 2024

    Olga Zaghen, Antonio Longa, Steve Azzolin, Lev Telyatnikov, Andrea Passerini, et al. Sheaf diffusion goes nonlinear: Enhancing gnns with adaptive sheaf laplacians.PROCEEDINGS OF MACHINE LEARNING RESEARCH, 251, 2024

    2024

  5. [13]

    A comprehensive survey on graph neural networks.IEEE Transactions on Neural Networks and Learning Systems, 32(1):4–24, 2020

    Zonghan Wu, Shirui Pan, Fengwen Chen, Guodong Long, Chengqi Zhang, and S Yu Philip. A comprehensive survey on graph neural networks.IEEE Transactions on Neural Networks and Learning Systems, 32(1):4–24, 2020

    2020

  6. [14]

    Dynamic graph cnn for learning on point clouds.Acm Transactions On Graphics (tog), 38(5):1–12, 2019

    Yue Wang, Yongbin Sun, Ziwei Liu, Sanjay E Sarma, Michael M Bronstein, and Justin M Solomon. Dynamic graph cnn for learning on point clouds.Acm Transactions On Graphics (tog), 38(5):1–12, 2019

    2019

  7. [15]

    Gated graph sequence neural networks

    Yujia Li, Richard Zemel, Marc Brockschmidt, and Daniel Tarlow. Gated graph sequence neural networks. InProceedings of ICLR’16, April 2016

    2016

  8. [16]

    Edge directionality improves learning on heterophilic graphs

    Emanuele Rossi, Bertrand Charpentier, Francesco Di Giovanni, Fabrizio Frasca, Stephan Günnemann, and Michael M Bronstein. Edge directionality improves learning on heterophilic graphs. InLearning on graphs conference, pages 25–1. PMLR, 2024. 10

    2024

  9. [17]

    Thomas. N. Kipf and Max Welling. Semi-supervised classification with graph convolutional net- works. In5th International Conference on Learning Representations, ICLR 2017 - Conference Track Proceedings, 2017

    2017

  10. [18]

    Rosenblum, and Andrew Lim

    Zekun Tong, Yuxuan Liang, Changsheng Sun, David S. Rosenblum, and Andrew Lim. Directed graph convolutional network, 2020

    2020

  11. [19]

    Rosenblum, and Andrew Lim

    Zekun Tong, Yuxuan Liang, Changsheng Sun, Xinke Li, David S. Rosenblum, and Andrew Lim. Digraph inception convolutional networks.Advances in Neural Information Processing Systems, 2020-December(NeurIPS):1–12, 2020

    2020

  12. [20]

    Fluxes, Laplacians, and Kasteleyn’s theorem

    Elliott H Lieb and Michael Loss. Fluxes, Laplacians, and Kasteleyn’s theorem. InStatistical Mechanics, pages 457–483. Springer, 1993

    1993

  13. [21]

    Sigmanet: One laplacian to rule them all

    Stefano Fiorini, Stefano Coniglio, Michele Ciavotta, and Enza Messina. Sigmanet: One laplacian to rule them all. InProceedings of the AAAI Conference on Artificial Intelligence, pages 7568–7576, 2023

    2023

  14. [22]

    smgc: A complex-valued graph convolutional network via magnetic laplacian for directed graphs, 2021

    Jie Zhang, Bo Hui, Po-Wei Harn, Min-Te Sun, and Wei-Shinn Ku. smgc: A complex-valued graph convolutional network via magnetic laplacian for directed graphs, 2021

    2021

  15. [23]

    Sheaf neural networks.arXiv preprint arXiv:2012.06333, 2020

    Jakob Hansen and Thomas Gebhart. Sheaf neural networks.arXiv preprint arXiv:2012.06333, 2020

    2012 arXiv

  16. [24]

    Let there be direction in hypergraph neural networks.Transactions on Machine Learning Research, 2024

    Stefano Fiorini, Stefano Coniglio, Michele Ciavotta, and Alessio Del Bue. Let there be direction in hypergraph neural networks.Transactions on Machine Learning Research, 2024

    2024

  17. [25]

    Geom-gcn: Geometric graph convolutional networks.arXiv preprint arXiv:2002.05287, 2020

    Hongbin Pei, Bingzhe Wei, Kevin Chen-Chuan Chang, Yu Lei, and Bo Yang. Geom-gcn: Geometric graph convolutional networks.arXiv preprint arXiv:2002.05287, 2020

    2002 arXiv

  18. [26]

    Two sides of the same coin: Heterophily and oversmoothing in graph convolutional neural networks

    Yujun Yan, Milad Hashemi, Kevin Swersky, Yaoqing Yang, and Danai Koutra. Two sides of the same coin: Heterophily and oversmoothing in graph convolutional neural networks. In2022 IEEE International Conference on Data Mining (ICDM), pages 1287–1292. IEEE, 2022

    2022

  19. [27]

    A critical look at the evaluation of gnns under heterophily: Are we really making progress? InThe Eleventh International Conference on Learning Representations, 2023

    Oleg Platonov, Denis Kuznedelev, Michael Diskin, Artem Babenko, and Liudmila Prokhorenkova. A critical look at the evaluation of gnns under heterophily: Are we really making progress? InThe Eleventh International Conference on Learning Representations, 2023

    2023

  20. [28]

    Graph attention networks

    Petar Veliˇckovi´c, Guillem Cucurull, Arantxa Casanova, Adriana Romero, Pietro Liò, and Yoshua Bengio. Graph attention networks. InInternational Conference on Learning Representations, 2018

    2018

  21. [29]

    Inductive representation learning on large graphs.Advances in neural information processing systems, 30:1–11, 2017

    Will Hamilton, Zhitao Ying, and Jure Leskovec. Inductive representation learning on large graphs.Advances in neural information processing systems, 30:1–11, 2017

    2017

  22. [30]

    Be- yond homophily in graph neural networks: Current limitations and effective designs.Advances in neural information processing systems, 33:7793–7804, 2020

    Jiong Zhu, Yujun Yan, Lingxiao Zhao, Mark Heimann, Leman Akoglu, and Danai Koutra. Be- yond homophily in graph neural networks: Current limitations and effective designs.Advances in neural information processing systems, 33:7793–7804, 2020

    2020

  23. [31]

    Adaptive universal generalized pagerank graph neural network

    Eli Chien, Jianhao Peng, Pan Li, and Olgica Milenkovic. Adaptive universal generalized pagerank graph neural network. InInternational Conference on Learning Representations, 2021

    2021

  24. [32]

    Beyond low-frequency information in graph convolutional networks

    Deyu Bo, Xiao Wang, Chuan Shi, and Huawei Shen. Beyond low-frequency information in graph convolutional networks. InProceedings of the AAAI conference on artificial intelligence, volume 35, pages 3950–3957, 2021

    2021

  25. [33]

    Mixhop: Higher-order graph convolutional architectures via sparsified neighborhood mixing

    Sami Abu-El-Haija, Bryan Perozzi, Amol Kapoor, Nazanin Alipourfard, Kristina Lerman, Hrayr Harutyunyan, Greg Ver Steeg, and Aram Galstyan. Mixhop: Higher-order graph convolutional architectures via sparsified neighborhood mixing. Ininternational conference on machine learning,...

    2019

  26. [34]

    Simple and deep graph convolutional networks

    Ming Chen, Zhewei Wei, Zengfeng Huang, Bolin Ding, and Yaliang Li. Simple and deep graph convolutional networks. InInternational conference on machine learning, pages 1725–1735. PMLR, 2020. 11 A Licensing The Cora, Citeseer, and PubMed datasets are available at https://linqs.o...

    2020

  27. [35]

    Thus, L eF ⪰0

    Since ||eδx||2 2 is a norm, ||eδx||2 2 ≥0 holds for all x∈C , thus implying L eF ⪰0 for all x∗ ∈C . Thus, L eF ⪰0 . By following the same arguments using eδ′ := eδD− 1 2 , one can show thatL eF N ⪰0as well. Theorem 2.L eFN ⪯2I. Proof. Let Q eF N :=D − 1 2 eδ∗eδD− 1 2 for the c...

    2018

  28. [36]

    An Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz with 380 GB RAM, equipped with an NVIDIA Ampere A100 40GB

  29. [37]

    Model Settings.We trained every learning model considered in this paper for up to 1000 epochs with early stops of 200

    A 12th Gen Intel(R) Core(TM) i9-12900KF CPU @ 3.20GHz CPU with 64 GB RAM, equipped with an NVIDIA RTX 4090 GPU. Model Settings.We trained every learning model considered in this paper for up to 1000 epochs with early stops of 200. We adopted a learning rate of {1·10 −2,2·10 −2...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.