pith. sign in

arxiv: 2606.23517 · v1 · pith:I7ABHPYXnew · submitted 2026-06-22 · 💻 cs.LG · stat.ML

Collapsed Effective Operators for Higher-order Structures

Maximilian Krahn , Lennart Bastian , Vikas Garg , Bj\"orn Schuller , Tolga Birdal This is my paper

Pith reviewed 2026-06-26 09:07 UTC · model grok-4.3

classification 💻 cs.LG stat.ML
keywords higher-order structuresSchur complementgraded Laplacianspectral operatorsspectral clusteringgraph signal processingtopological neural networkseffective operators
0
0 comments X

The pith

Higher-order structures collapse into one effective vertex operator through Schur complementation of a graded Laplacian.

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

The paper establishes that higher-order topological features can be condensed into a single operator acting only on vertices, eliminating the need for separate rank handling and ad hoc fusion. Schur complementation applied to the graded Laplacian produces a generally dense operator that captures long-range interactions mediated by the full topology. This operator remains positive semi-definite and is bounded above in spectrum by the rank-0 Hodge Laplacian, which corresponds to a reduction in system energy when higher-order connectivity is present. The construction applies to arbitrary higher-order constructs and yields measurable gains in spectral clustering, signal smoothing, and topological positional encodings inside neural networks.

Core claim

We introduce Collapsed Effective Operators, which condense higher-order degrees of freedom into a single vertex-level operator via Schur complementation of a graded Laplacian. This yields a (generally dense) operator that encodes long-range interactions mediated by topology and is applicable to arbitrary higher-order constructs. We show it preserves positive semi-definiteness with a spectral upper bound relative to the rank-0 Hodge Laplacian, effectively lowering system energy under higher-order connectivity.

What carries the argument

Schur complementation of the graded Laplacian, which reduces all higher-order degrees of freedom to a single vertex-level operator while retaining long-range topological interactions.

If this is right

  • Spectral clustering and signal smoothing can now incorporate higher-order information without choosing how to recombine separate ranks at the vertex level.
  • Neural network positional encodings can directly embed topological features derived from the full graded structure.
  • Any higher-order construct gains an effective low-energy description at the vertex level that is guaranteed to respect the original positive semi-definiteness.
  • Adding higher-order connectivity is guaranteed to lower or maintain the quadratic form energy relative to the ordinary graph Laplacian.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same reduction technique might be applied to other matrix pencils arising in algebraic topology to obtain effective low-dimensional models.
  • In large-scale networks the dense collapsed operator could be approximated by sparse or low-rank surrogates while preserving the spectral bound.
  • The energy-lowering property suggests the operator may serve as a natural regularizer when higher-order data is noisy or incomplete.

Load-bearing premise

The graded Laplacian is well-defined for arbitrary higher-order constructs and the Schur complement can be formed such that the resulting vertex operator inherits the claimed spectral properties without further restrictions on the topology or the choice of grading.

What would settle it

A concrete higher-order complex where the collapsed operator either loses positive semi-definiteness or exceeds the largest eigenvalue of the rank-0 Hodge Laplacian would disprove the spectral preservation claim.

Figures

Figures reproduced from arXiv: 2606.23517 by Bj\"orn Schuller, Lennart Bastian, Maximilian Krahn, Tolga Birdal, Vikas Garg.

Figure 1
Figure 1. Figure 1: Our Collapsed effective operators reveal secondary-structure-aware spectral modes. On protein 1A0C (437 residues) from Topotein (Wang et al., 2025), we color residues by the leading spectral embedding (top-3 joint-PCA components of the leading eigenvectors) of our collapsed operator S versus the graph Laplacian L 𝐺, aligned so matching modes share colors. Left to right: (I) the secondary structure, shown a… view at source ↗
Figure 2
Figure 2. Figure 2: Topological domains of increasing flexibility. (a) Graph: pairwise edges. (b) Hypergraph: edges of arbitrary size, all flat (no rank). (c) Simplicial complex: higher-order simplices, each determined by its faces. (d) Cell complex: general cells whose boundaries need not be simplices. (e) Combinatorial complex: possibly overlapping cells with explicit rank. et al., 2022) [PITH_FULL_IMAGE:figures/full_fig_p… view at source ↗
Figure 3
Figure 3. Figure 3: Unsupervised protein segmentation on 1LYZ from the Topotein benchmark (Wang et al., 2025). Left: Ground truth. Mid￾dle: The standard Graph Laplacian (46.9%) fails to capture inter￾leaved motifs. Right: Our effective operator (70.9%) successfully recovers these non-local structures, significantly outperforming the baseline. spiral backbone conformations; 𝛽-sheets (E), consisting of extended parallel or anti… view at source ↗
Figure 4
Figure 4. Figure 4: Rank completion for a skipped containment relation in a CC. A direct relation 𝑥 ≺ 𝑦 with a rank gap is replaced by a chain through formal intermediate elements, producing adjacent-rank cover relations for the incidence matrices used by the Graded Laplacian. • Generalized Connectivity: Unlike the homological case, this setting allows for I𝑘−1I𝑘 ≠ 0. This non-vanishing product captures hierarchical co-occurr… view at source ↗
read the original abstract

Higher-order structures are powerful relational modeling tools, yet existing spectral operators decompose the topology into separate ranks, leaving practitioners to fuse the information back to vertices through ad hoc choices. We introduce Collapsed Effective Operators, which condense higher-order degrees of freedom into a single vertex-level operator via Schur complementation of a graded Laplacian. This yields a (generally dense) operator that encodes long-range interactions mediated by topology and is applicable to arbitrary higher-order constructs. We show it preserves positive semi-definiteness with a spectral upper bound relative to the rank-0 Hodge Laplacian, effectively lowering system energy under higher-order connectivity. Empirically, our operator improves spectral clustering, signal smoothing, and enables the inclusion of topological features in neural network architectures via positional encoding. The project page can be found http://circle-group.github.io/research/CollapsedEffectiveOperators

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 0 minor

Summary. The paper introduces Collapsed Effective Operators that condense higher-order degrees of freedom into a single vertex-level operator via Schur complementation of a graded Laplacian. This produces a generally dense operator encoding long-range topological interactions applicable to arbitrary higher-order constructs. The central claims are that the resulting operator preserves positive semi-definiteness and admits a spectral upper bound relative to the rank-0 Hodge Laplacian (thereby lowering system energy under higher-order connectivity), with empirical demonstrations on spectral clustering, signal smoothing, and neural-network positional encodings.

Significance. If the claimed PSD preservation and spectral upper bound can be established with explicit conditions, the construction would supply a systematic, non-ad-hoc route for incorporating higher-order topology directly into vertex-level spectral operators, which could strengthen topological signal processing and graph neural network architectures.

major comments (1)
  1. [Abstract] Abstract: the claim that Schur complementation of an arbitrary graded Laplacian 'preserves positive semi-definiteness with a spectral upper bound relative to the rank-0 Hodge Laplacian' is asserted without derivation, block-structure assumptions, or error analysis. The inheritance of these properties is not automatic and depends on the definiteness of the eliminated blocks and the signs of the off-diagonal couplings; without explicit conditions the general applicability asserted in the abstract is unsupported.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful reading and for highlighting the need for explicit conditions in the abstract. We address the concern below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that Schur complementation of an arbitrary graded Laplacian 'preserves positive semi-definiteness with a spectral upper bound relative to the rank-0 Hodge Laplacian' is asserted without derivation, block-structure assumptions, or error analysis. The inheritance of these properties is not automatic and depends on the definiteness of the eliminated blocks and the signs of the off-diagonal couplings; without explicit conditions the general applicability asserted in the abstract is unsupported.

    Authors: We agree that the abstract, as a concise summary, does not spell out the block assumptions or derivation. The main text (Section 3) establishes PSD preservation via the Schur complement formula when the graded Laplacian is PSD and the eliminated block (higher-order simplices) is positive definite; the spectral upper bound follows from the variational min-max theorem applied to the quadratic form of the effective operator. We will revise the abstract to state the result under the explicit condition that the higher-order blocks are positive definite, and we will add a forward reference to the relevant theorem. No error analysis is claimed beyond the exact Schur complement identity. revision: yes

Circularity Check

0 steps flagged

No circularity: operator defined directly via Schur complement with independent spectral claims

full rationale

The paper defines Collapsed Effective Operators explicitly as the result of Schur complementation applied to a graded Laplacian assembled from higher-order structures. The claimed preservation of positive semi-definiteness and the spectral upper bound relative to the rank-0 Hodge Laplacian are presented as properties shown from this construction, without any reduction of the result to fitted parameters, self-referential definitions, or load-bearing self-citations. No equations equate the output operator back to its inputs by construction, and the abstract supplies no ansatz or uniqueness theorem imported from prior author work. The derivation chain is therefore self-contained as a direct algebraic construction.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 1 invented entities

The method rests on standard properties of Schur complements and the definition of graded Laplacians from higher-order topology; no free parameters or new entities with independent evidence are introduced in the abstract.

axioms (2)
  • domain assumption Graded Laplacian is defined and well-posed for arbitrary higher-order constructs
    Invoked when stating the operator is applicable to arbitrary higher-order constructs.
  • standard math Schur complement of a PSD matrix remains PSD under the grading used
    Required for the claimed preservation of positive semi-definiteness.
invented entities (1)
  • Collapsed Effective Operator no independent evidence
    purpose: Condenses higher-order degrees of freedom into a vertex-level operator
    The central new object defined by the Schur complement construction.

pith-pipeline@v0.9.1-grok · 5680 in / 1363 out tokens · 28371 ms · 2026-06-26T09:07:53.503040+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

182 extracted references · 4 linked inside Pith

  1. [1]

    FirstName LastName , title =

  2. [2]

    FirstName Alpher , title =

  3. [3]

    Journal of Foo , volume = 13, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe , title =. Journal of Foo , volume = 13, number = 1, pages =

  4. [4]

    Journal of Foo , volume = 14, number = 1, pages =

    FirstName Alpher and FirstName Fotheringham-Smythe and FirstName Gamow , title =. Journal of Foo , volume = 14, number = 1, pages =

  5. [5]

    FirstName Alpher and FirstName Gamow , title =

  6. [6]

    Linear algebra and its applications , volume=

    Laplacian matrices of graphs: a survey , author=. Linear algebra and its applications , volume=. 1994 , publisher=

    1994

  7. [7]

    arXiv preprint arXiv:0712.1074 , year=

    On sumsets of dissociated sets , author=. arXiv preprint arXiv:0712.1074 , year=

    Pith/arXiv arXiv

  8. [8]

    Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

    Deepwalk: Online learning of social representations , author=. Proceedings of the 20th ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

  9. [9]

    Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

    node2vec: Scalable feature learning for networks , author=. Proceedings of the 22nd ACM SIGKDD international conference on Knowledge discovery and data mining , pages=

  10. [10]

    Proceedings of the 24th international conference on world wide web , pages=

    Line: Large-scale information network embedding , author=. Proceedings of the 24th international conference on world wide web , pages=

  11. [11]

    Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining , pages=

    struc2vec: Learning node representations from structural identity , author=. Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining , pages=

  12. [12]

    Computer graphics forum , volume=

    A concise and provably informative multi-scale signature based on heat diffusion , author=. Computer graphics forum , volume=. 2009 , organization=

    2009

  13. [13]

    2011 IEEE international conference on computer vision workshops (ICCV workshops) , pages=

    The wave kernel signature: A quantum mechanical approach to shape analysis , author=. 2011 IEEE international conference on computer vision workshops (ICCV workshops) , pages=. 2011 , organization=

    2011

  14. [14]

    Weisfeiler and lehman go cellular: Cw networks , author=

  15. [15]

    2021 , organization=

    Weisfeiler and lehman go topological: Message passing simplicial networks , author=. 2021 , organization=

    2021

  16. [16]

    arXiv preprint arXiv:2010.00743 , year=

    Cell complex neural networks , author=. arXiv preprint arXiv:2010.00743 , year=

    arXiv 2010

  17. [18]

    arXiv preprint arXiv:2206.00606 , year=

    Topological deep learning: Going beyond graph data , author=. arXiv preprint arXiv:2206.00606 , year=

    arXiv

  18. [19]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Higher-order graph convolutional network with flower-petals laplacians on simplicial complexes , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  19. [20]

    2018 , publisher=

    Geometric and topological inference , author=. 2018 , publisher=

    2018

  20. [21]

    Frontiers in artificial intelligence , volume=

    An introduction to topological data analysis: fundamental and practical aspects for data scientists , author=. Frontiers in artificial intelligence , volume=. 2021 , publisher=

    2021

  21. [22]

    Advances in applied and computational topology , volume=

    Topological data analysis , author=. Advances in applied and computational topology , volume=. 2012 , publisher=

    2012

  22. [23]

    ICLR 2025 Workshop on Generative and Experimental Perspectives for Biomolecular Design , year=

    Higher-Order Molecular Learning: The Cellular Transformer , author=. ICLR 2025 Workshop on Generative and Experimental Perspectives for Biomolecular Design , year=

    2025

  23. [24]

    2023 57th Asilomar Conference on Signals, Systems, and Computers , pages=

    Combinatorial complexes: bridging the gap between cell complexes and hypergraphs , author=. 2023 57th Asilomar Conference on Signals, Systems, and Computers , pages=. 2023 , organization=

    2023

  24. [25]

    Proceedings of machine learning research , volume=

    Position: Topological deep learning is the new frontier for relational learning , author=. Proceedings of machine learning research , volume=

  25. [26]

    Topological Graph Neural Networks , author=

  26. [27]

    Advances in neural information processing systems , volume=

    Topological relational learning on graphs , author=. Advances in neural information processing systems , volume=

  27. [28]

    arXiv preprint arXiv:2302.09826 , year=

    On the expressivity of persistent homology in graph learning , author=. arXiv preprint arXiv:2302.09826 , year=

    arXiv

  28. [29]

    Topological neural networks go persistent, equivariant, and continuous , author=

  29. [30]

    arXiv preprint arXiv:2409.08217 , year=

    CliquePH: Higher-Order Information for Graph Neural Networks through Persistent Homology on Clique Graphs , author=. arXiv preprint arXiv:2409.08217 , year=

    arXiv

  30. [31]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Weisfeiler and leman go neural: Higher-order graph neural networks , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  31. [32]

    2019 , url=

    How Powerful are Graph Neural Networks? , author=. 2019 , url=

    2019

  32. [33]

    nti, Series , volume=

    The reduction of a graph to canonical form and the algebra which appears therein , author=. nti, Series , volume=

  33. [34]

    International Conference on Learning Representations , year=

    Invariant and Equivariant Graph Networks , author=. International Conference on Learning Representations , year=

  34. [35]

    arXiv preprint arXiv:2010.01179 , year=

    The surprising power of graph neural networks with random node initialization , author=. arXiv preprint arXiv:2010.01179 , year=

    arXiv 2010

  35. [36]

    IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=

    Improving graph neural network expressivity via subgraph isomorphism counting , author=. IEEE Transactions on Pattern Analysis and Machine Intelligence , volume=. 2022 , publisher=

    2022

  36. [37]

    Advances in neural information processing systems , volume=

    Provably powerful graph networks , author=. Advances in neural information processing systems , volume=

  37. [38]

    2023 , organization=

    Equivariant polynomials for graph neural networks , author=. 2023 , organization=

    2023

  38. [39]

    International Conference on Learning Representations , year=

    Equivariant subgraph aggregation networks , author=. International Conference on Learning Representations , year=

  39. [40]

    Advances in Neural Information Processing Systems , volume=

    Understanding and extending subgraph gnns by rethinking their symmetries , author=. Advances in Neural Information Processing Systems , volume=

  40. [41]

    2023 , organization=

    A complete expressiveness hierarchy for subgraph gnns via subgraph weisfeiler-lehman tests , author=. 2023 , organization=

    2023

  41. [42]

    Advances in Neural Information Processing Systems , volume=

    Nested graph neural networks , author=. Advances in Neural Information Processing Systems , volume=

  42. [43]

    Advances in Neural Information Processing Systems , volume=

    Reconstruction for powerful graph representations , author=. Advances in Neural Information Processing Systems , volume=

  43. [44]

    A Flexible, Equivariant Framework for Subgraph

    Guy Bar-Shalom and Yam Eitan and Fabrizio Frasca and Haggai Maron , booktitle=. A Flexible, Equivariant Framework for Subgraph. 2024 , url=

    2024

  44. [45]

    Topological, Algebraic and Geometric Learning Workshops 2022 , pages=

    A Topological characterisation of Weisfeiler-Leman equivalence classes , author=. Topological, Algebraic and Geometric Learning Workshops 2022 , pages=. 2022 , organization=

    2022

  45. [46]

    The International Congress of Mathematicians , pages=

    Theory of graph neural networks: Representation and learning , author=. The International Congress of Mathematicians , pages=

  46. [47]

    Journal of Machine Learning Research , volume=

    Weisfeiler and leman go machine learning: The story so far , author=. Journal of Machine Learning Research , volume=

  47. [48]

    IEEE Transactions on Knowledge and Data Engineering , year=

    The expressive power of graph neural networks: A survey , author=. IEEE Transactions on Knowledge and Data Engineering , year=

  48. [49]

    Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

    Learning structural node embeddings via diffusion wavelets , author=. Proceedings of the 24th ACM SIGKDD international conference on knowledge discovery & data mining , pages=

  49. [50]

    , author=

    Graph laplacians and their convergence on random neighborhood graphs. , author=. Journal of Machine Learning Research , volume=

  50. [51]

    Computer graphics forum , volume=

    Global intrinsic symmetries of shapes , author=. Computer graphics forum , volume=. 2008 , organization=

    2008

  51. [52]

    1997 , publisher=

    Spectral graph theory , author=. 1997 , publisher=

    1997

  52. [53]

    Advances in neural information processing systems , volume=

    On spectral clustering: Analysis and an algorithm , author=. Advances in neural information processing systems , volume=

  53. [54]

    Neural computation , volume=

    Laplacian eigenmaps for dimensionality reduction and data representation , author=. Neural computation , volume=. 2003 , publisher=

    2003

  54. [55]

    Applied and computational harmonic analysis , volume=

    Diffusion maps , author=. Applied and computational harmonic analysis , volume=. 2006 , publisher=

    2006

  55. [56]

    International Conference on Learning Representations , year=

    Semi-Supervised Classification with Graph Convolutional Networks , author=. International Conference on Learning Representations , year=

  56. [57]

    arXiv preprint arXiv:2304.10031 , year=

    Architectures of topological deep learning: A survey of message-passing topological neural networks , author=. arXiv preprint arXiv:2304.10031 , year=

    arXiv

  57. [58]

    The Thirteenth International Conference on Learning Representations , year=

    Topological Blindspots: Understanding and Extending Topological Deep Learning Through the Lens of Expressivity , author=. The Thirteenth International Conference on Learning Representations , year=

  58. [59]

    Theory and Applications of Graphs , volume=

    The gamma-Signless Laplacian Adjacency Matrix of Mixed Graphs , author=. Theory and Applications of Graphs , volume=

  59. [60]

    2022 , organization=

    Convergence of invariant graph networks , author=. 2022 , organization=

    2022

  60. [61]

    Advances in neural information processing systems , volume=

    Mlp-mixer: An all-mlp architecture for vision , author=. Advances in neural information processing systems , volume=

  61. [62]

    International Conference on Scale Space and Variational Methods in Computer Vision , pages=

    Graph laplacian for semi-supervised learning , author=. International Conference on Scale Space and Variational Methods in Computer Vision , pages=. 2023 , organization=

    2023

  62. [63]

    Machine Learning for Structural Biology Workshop at NeurIPS 2024 , year=

    Higher-Order Message Passing for Glycan Representation Learning , author=. Machine Learning for Structural Biology Workshop at NeurIPS 2024 , year=

    2024

  63. [64]

    Neural networks , volume=

    Multilayer feedforward networks are universal approximators , author=. Neural networks , volume=. 1989 , publisher=

    1989

  64. [65]

    International Conference on Learning Representations , year=

    Are transformers universal approximators of sequence-to-sequence functions? , author=. International Conference on Learning Representations , year=

  65. [66]

    Topology , volume=

    Witten--Morse theory for cell complexes , author=. Topology , volume=. 1998 , publisher=

    1998

  66. [67]

    Journal of Data-centric Machine Learning Research , year=

    TopoBench: A Framework for Benchmarking Topological Deep Learning , author=. Journal of Data-centric Machine Learning Research , year=

  67. [68]

    The Thirteenth International Conference on Learning Representations , year=

    E(n) Equivariant Topological Neural Networks , author=. The Thirteenth International Conference on Learning Representations , year=

  68. [69]

    2021 , organization=

    Principled simplicial neural networks for trajectory prediction , author=. 2021 , organization=

    2021

  69. [70]

    arXiv preprint arXiv:2010.03633 , year=

    Simplicial neural networks , author=. arXiv preprint arXiv:2010.03633 , year=

    arXiv 2010

  70. [71]

    IEEE Transactions on Signal and Information Processing over Networks , year=

    Generalized simplicial attention neural networks , author=. IEEE Transactions on Signal and Information Processing over Networks , year=

  71. [72]

    arXiv preprint arXiv:2312.08515 , year=

    Simplicial representation learning with neural k -forms , author=. arXiv preprint arXiv:2312.08515 , year=

    arXiv

  72. [73]

    ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=

    Higher-order topological directionality and directed simplicial neural networks , author=. ICASSP 2025-2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP) , pages=. 2025 , organization=

    2025

  73. [74]

    2010 IEEE computer society conference on computer vision and pattern recognition , pages=

    Scale-invariant heat kernel signatures for non-rigid shape recognition , author=. 2010 IEEE computer society conference on computer vision and pattern recognition , pages=. 2010 , organization=

    2010

  74. [75]

    Proceedings of the ACM workshop on 3D object retrieval , pages=

    Volumetric heat kernel signatures , author=. Proceedings of the ACM workshop on 3D object retrieval , pages=

  75. [76]

    Computer Graphics Forum , volume=

    One point isometric matching with the heat kernel , author=. Computer Graphics Forum , volume=. 2010 , organization=

    2010

  76. [77]

    Joint IAPR International Workshops on Statistical Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition (SSPR) , pages=

    A novel graph kernel based on the Wasserstein distance and spectral signatures , author=. Joint IAPR International Workshops on Statistical Techniques in Pattern Recognition (SPR) and Structural and Syntactic Pattern Recognition (SSPR) , pages=. 2022 , organization=

    2022

  77. [78]

    International conference on machine learning , pages=

    p -Laplacian Based Graph Neural Networks , author=. International conference on machine learning , pages=. 2022 , organization=

    2022

  78. [79]

    arXiv preprint arXiv:2208.01853 , year=

    Robust graph neural networks using weighted graph laplacian , author=. arXiv preprint arXiv:2208.01853 , year=

    arXiv

  79. [80]

    arXiv preprint arXiv:2406.12841 , year=

    Demystifying Higher-Order Graph Neural Networks , author=. arXiv preprint arXiv:2406.12841 , year=

    arXiv

  80. [81]

    Learning on Graphs Conference , pages=

    Representing edge flows on graphs via sparse cell complexes , author=. Learning on Graphs Conference , pages=. 2024 , organization=

    2024

Showing first 80 references.