pith. machine review for the scientific record. sign in

arxiv: 2604.24672 · v1 · submitted 2026-04-27 · 💻 cs.LG · math.AT

Recognition: unknown

A Functorial Formulation of Neighborhood Aggregating Deep Learning

Sun Woo Park , Yun Young Choi , U Jin Choi , Youngho Woo
Authors on Pith no claims yet

Pith reviewed 2026-05-08 04:13 UTC · model grok-4.3

classification 💻 cs.LG math.AT
keywords presheavescopresheavessheaf obstructionsneighborhood aggregationconvolutional neural networksmessage passing networkstopological spacesfunctorial formulation
0
0 comments X

The pith

Neighborhood aggregation in CNNs and message-passing networks admits a presheaf interpretation whose obstructions to being sheaves explain observed empirical limitations.

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

The paper models the neighborhood aggregation steps of convolutional neural networks and message-passing networks as presheaves or copresheaves whose sections are continuous functions defined on a topological space. This categorical interpretation supplies a heuristic that connects the failure of these structures to satisfy the full sheaf or copresheaf axioms to concrete performance shortcomings seen in practice. A reader would care because the approach offers a unified, topology-based account of why local aggregation operations sometimes produce global inconsistencies or restricted expressivity in these architectures.

Core claim

We provide a mathematical interpretation of convolutional (or message passing) neural networks by using presheaves and copresheaves of the set of continuous functions over a topological space. Based on this interpretation, we formulate a theoretical heuristic which elaborates a number of empirical limitations of these neural networks by using obstructions on such sets of continuous functions over a topological space to be sheaves or copresheaves.

What carries the argument

Presheaves and copresheaves of continuous functions on topological spaces that model neighborhood aggregation, with obstructions to sheaf or copresheaf conditions as the explanatory device.

If this is right

  • Empirical shortcomings of these networks can be classified according to which sheaf axioms are violated by their aggregation rules.
  • The same presheaf framework applies uniformly to both convolutional and graph-based neighborhood operations.
  • Obstructions arising from local-to-global inconsistency provide a systematic way to predict when aggregation will lose information or produce artifacts.

Where Pith is reading between the lines

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

  • If the correspondence is tight, one could design aggregation operators that explicitly enforce selected sheaf conditions to reduce the identified limitations.
  • The approach suggests checking whether other aggregation-based models outside CNNs and GNNs exhibit analogous obstructions when cast in the same presheaf language.
  • Explicit computation of sheaf cohomology on small example graphs or grids could quantify the size of the obstructions and correlate them with measured performance gaps.

Load-bearing premise

The presheaf or copresheaf model of neighborhood aggregation is faithful enough that its sheaf-theoretic obstructions actually correspond to the empirical limitations of the networks.

What would settle it

A concrete counterexample in which a CNN or message-passing network violates a specific sheaf gluing condition yet does not exhibit the corresponding empirical limitation, or vice versa.

Figures

Figures reproduced from arXiv: 2604.24672 by Sun Woo Park, U Jin Choi, Youngho Woo, Yun Young Choi.

Figure 1
Figure 1. Figure 1: Construction of a pushforward of constant sheaves / cosheaves over discrete sets view at source ↗
Figure 2
Figure 2. Figure 2: Construction of a functor describing the presheaf / copresheaf of continuous functions induced from the sheaf iA,l over a topological space X and its universal cover Xf. Let us regard the functor iA,l as the sheaf of real vector spaces over X. The inclusion of open subsets U → V induces a projection map of real vector spaces. resV,U : iA,l(V ) → iA,l(U) (y1, y2, · · · , yP xi∈V li ) 7→ (y1, y2, · · · , yP … view at source ↗
Figure 3
Figure 3. Figure 3: Definition of the restriction map resX,U view at source ↗
Figure 4
Figure 4. Figure 4: Definition of the inclusion map iU,X is not surjective. Consider the function f(y1, · · · , yPN i=1 li ) =   PN i=1 Y li j=1 yj , · · · , PN i=1 Y li j=1 yj   (21) Then there doesn’t exist a collection of functions {gi : R li → R k} N i=1 such that f = P α gα ◦ resUα,U . Indeed, for each i-th coordinate of f, gi ̸= Q PN i=1 li j=1 yj . The presheaf C 0 (iA,l, R k ) If we consider Sxi as the skyscrape… view at source ↗
Figure 5
Figure 5. Figure 5: Summation by inclusion maps, which defines the surjectivity condition of the cosheaf axiom view at source ↗
Figure 6
Figure 6. Figure 6: An example of a layer of a convolutional (or message passing) neural network Remark 3.9. The definition provided above generalizes the sum-decomposition of continuous functions f : R n → R proposed in [52] and [49]. A function f : R n → R is said to be sum￾decomposable if there exist functions ϕ : R → Z and ρ : Z → R for some topological space Z such that f(x1, · · · , xn) = ρ( Xn i=1 ϕ(xi)) (43) Example 3… view at source ↗
Figure 7
Figure 7. Figure 7: Presheaf: Does not satisfy the locality condition view at source ↗
Figure 8
Figure 8. Figure 8: Presheaf: Satisfies the gluing condition whereas for any vector v ∈ R k0×N and p > 0, view at source ↗
Figure 9
Figure 9. Figure 9: Copresheaf: Does not satisfy the surjectivity condition view at source ↗
Figure 10
Figure 10. Figure 10: Copresheaf: Satisfies the gluing condition Definition 3.19 ([24], Proposition-Definition 2.1.2). Let F be a presheaf over X. Define a contravariant functor F + from the category of open subsets of X to the category of abelian groups as follows. F +(U) := {s : U → G x∈U Fx such that (∗)} (62) where the conditions (∗) are: (1) For each x ∈ U, s(x) ∈ Fx (2) For each x ∈ U, there exists an open neighborhood V… view at source ↗
Figure 11
Figure 11. Figure 11: Sheafification: There are no obstructions imposed from topological characteristics of X in obtaining global sections of C 0,+(iA,l, R k ). As such, non-trivial geometric differences among topological spaces cannot be effectively detected from neighborhood aggregating discrete deep learning techniques. isomorphism tests, these representations are also obtained from the set of local sections on U k v , the … view at source ↗
read the original abstract

We provide a mathematical interpretation of convolutional (or message passing) neural networks by using presheaves and copresheaves of the set of continuous functions over a topological space. Based on this interpretation, we formulate a theoretical heuristic which elaborates a number of empirical limitations of these neural networks by using obstructions on such sets of continuous functions over a topological space to be sheaves or copresheaves.

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

2 major / 0 minor

Summary. The paper claims to provide a mathematical interpretation of convolutional and message-passing neural networks via presheaves and copresheaves on the set of continuous functions over a topological space, then formulates a heuristic that attributes empirical limitations of these networks (such as over-smoothing or restricted receptive fields) to obstructions preventing the structures from satisfying the sheaf or copresheaf axioms.

Significance. If the proposed presheaf/copresheaf interpretation can be equipped with explicit functors and shown to predict specific documented failures, the work would supply a categorical lens on neighborhood aggregation that could guide architecture design; the attempt to connect sheaf obstructions to learning limitations is a potentially useful direction, though currently unrealized.

major comments (2)
  1. [Abstract] Abstract: the central claim asserts that neighborhood aggregation in CNNs and message-passing networks can be modeled as (co)presheaves such that failures of the locality or gluing axioms directly reproduce cited empirical limitations, yet supplies neither an explicit functor from network layers to presheaf data nor a concrete example (e.g., how a graph convolution operator violates the gluing condition and thereby produces over-smoothing).
  2. [Main text] Main text (heuristic formulation): the obstructions are defined in terms of the same presheaf/copresheaf structures used to interpret the networks, creating a risk that the heuristic restates known limitations rather than deriving new, independently verifiable predictions; no independent test or counter-example is supplied to break the potential circularity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments and the recognition of the potential value in a categorical perspective on neighborhood aggregation. We respond point-by-point to the major comments below, indicating revisions where appropriate to strengthen the explicitness of the functorial construction and the heuristic.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim asserts that neighborhood aggregation in CNNs and message-passing networks can be modeled as (co)presheaves such that failures of the locality or gluing axioms directly reproduce cited empirical limitations, yet supplies neither an explicit functor from network layers to presheaf data nor a concrete example (e.g., how a graph convolution operator violates the gluing condition and thereby produces over-smoothing).

    Authors: We agree that greater explicitness would improve the presentation. The manuscript defines the presheaf of continuous functions on the underlying topological space, with neighborhood aggregation corresponding to the restriction morphisms of the (co)presheaf; this supplies the functorial mapping from layers to the categorical data. To address the request directly, the revised manuscript will include an explicit functor definition from a general neighborhood-aggregating operator to the associated presheaf/copresheaf and a concrete worked example (a graph convolution on a simple cycle graph) that isolates the gluing obstruction and links it to over-smoothing. revision: yes

  2. Referee: [Main text] Main text (heuristic formulation): the obstructions are defined in terms of the same presheaf/copresheaf structures used to interpret the networks, creating a risk that the heuristic restates known limitations rather than deriving new, independently verifiable predictions; no independent test or counter-example is supplied to break the potential circularity.

    Authors: The heuristic is not circular: the (co)presheaf model is fixed first as the mathematical interpretation of aggregation, after which the obstructions are derived strictly from the failure of the sheaf axioms (locality and gluing). These axiom failures are independent of any neural-network performance metric and are then connected heuristically to documented empirical behaviors. We acknowledge, however, that an independent verification would strengthen the claim. The revised version will therefore add a short section proposing a concrete, testable prediction (an architecture modification that enforces the gluing condition via auxiliary regularization) together with the expected effect on over-smoothing benchmarks. revision: partial

Circularity Check

0 steps flagged

No load-bearing circularity; interpretation and heuristic remain modeling choices without reduction to inputs by construction.

full rationale

The abstract states that the authors provide a presheaf/copresheaf interpretation of neighborhood aggregation and then formulate a heuristic that uses obstructions to sheaf/copresheaf axioms to elaborate empirical limitations. No equations, self-citations, or fitted parameters are quoted that would make any claimed prediction or obstruction equivalent to the input interpretation by definition. The correspondence between sheaf obstructions and limitations is presented as a heuristic rather than a derived equality or statistically forced output. The derivation chain is therefore self-contained as an act of mathematical modeling; no step reduces to renaming or fitting the same structure it claims to explain.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on standard category-theoretic definitions of presheaves, copresheaves, and sheaves applied to the set of continuous functions; no free parameters or new entities are introduced in the abstract.

axioms (2)
  • domain assumption Presheaves and copresheaves of continuous functions on a topological space can serve as a faithful model for neighborhood aggregation in convolutional and message-passing networks.
    Invoked in the opening sentence of the abstract as the basis for the mathematical interpretation.
  • ad hoc to paper Obstructions to the presheaf data forming a sheaf or copresheaf correspond to empirical limitations of the networks.
    This is the load-bearing step of the proposed heuristic; it is stated but not derived in the abstract.

pith-pipeline@v0.9.0 · 5355 in / 1369 out tokens · 22388 ms · 2026-05-08T04:13:26.321836+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

55 extracted references · 2 canonical work pages

  1. [1]

    Approximation theory of the MLP model in neural networks

    Allan Pinkus “Approximation theory of the MLP model in neural networks”. Acta Numerica. vol.8 pp.143–195 (1999)

    1999

  2. [2]

    A Topological characterisation of Weisfeiler-Leman equivalence classes

    Jacob Bamberger. “A Topological characterisation of Weisfeiler-Leman equivalence classes”. ICML Workshop: Topology, Algebra, and Geometry in Machine Learning. (2022)

    2022

  3. [3]

    Sheaf Neural Networks with Connection Laplacians

    Federico Barbero, Cristian Bodnar, Haitz Sáez de Ocáriz Borde, Michael Bronstein, Petar Veličković, and Pietro Liò. “Sheaf Neural Networks with Connection Laplacians”. ICML 2022 Workshop on Topology, Algebra, and Geometry in Machine Learning. (2022)

    2022

  4. [4]

    Sheaf attention networks

    Federico Barbero, Christian Bodnar, Haitz Sáez de Ocáriz Borde, and Pietro Liò. “Sheaf attention networks”. NeurIPS’22 Workshop on Symmetry and Geometry in Neural Representations (2022)

    2022

  5. [5]

    Heterogeneous Sheaf Neural Networks

    Luke Braithwaite, Alessio Borgi, Gabriele Onorato, Kristjan Tarantelli, Iulia Duta, Francesco Restuccia, Fabrizio Silvestri, and Pietro Liò. “Heterogeneous Sheaf Neural Networks”. Arxiv preprint (2024)

    2024

  6. [6]

    Neural Sheaf Diffusion: A Topological Perspective on Heterophily and Oversmoothing in GNNs

    Cristian Bodnar, Francesco Di Giovanni, Benjamin Paul Chamberlain, Pietro Lio, and Michael M. Bronstein. “Neural Sheaf Diffusion: A Topological Perspective on Heterophily and Oversmoothing in GNNs”. NeurIPS’22: Proceedings of the 36th International Conference on Neural Information Processing Systems. (2022)

    2022

  7. [7]

    Polynomial Neural Sheaf Diffusion: A Spectral Filtering Approach on Cellular Sheaves

    Alessio Borgi, Fabrizio Silvestri, and Pietro Liò. “Polynomial Neural Sheaf Diffusion: A Spectral Filtering Approach on Cellular Sheaves”. Arxiv Preprint. (2025)

    2025

  8. [8]

    Sheaf Theory

    Glen E. Bredon. “Sheaf Theory”. Graduate Texts in Mathematics, Springer. 279–448 (1997)

    1997

  9. [9]

    Perslay: a neural network layer for persistence diagrams and new graph topological signatures

    Mathieu Carriere, Frederic Chazal, Yuichi Ike, Theo Lacombe, Martin Royer, Yuhei Umeda. “Perslay: a neural network layer for persistence diagrams and new graph topological signatures”. Proceedings of the 23rd International Conference on Artificial Intelligence and Statistics (2020)

    2020

  10. [10]

    Neural Ordinary Differential Equations

    Ricky T.Q.Chen, Yulia Rubanova, Jesse Bettencourt, and David Duvenaud. “Neural Ordinary Differential Equations”. Advances in Neural Information Processing Systems. Vol. 31. (2019)

    2019

  11. [11]

    Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

    Kyunghyun Cho, Bart van Merrienboer, Caglar Culcehre, Dzmitry Bahdanau Fethi Bougares, Holger Schwenk, Yoshua Bengio. “Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation”. Empirical Methods in Natural Language Processing (EMNLP). pp.1724-1734. (2014)

    2014

  12. [12]

    Cycle to Clique (Cy2C) Graph Neural Network: A Sight to See beyond Neighborhood Aggregation

    Yunyoung Choi, Sun Woo Park, U Jin Choi, and Youngho Woo. “Cycle to Clique (Cy2C) Graph Neural Network: A Sight to See beyond Neighborhood Aggregation”. Preprint, Submitted. (2022)

    2022

  13. [13]

    Equivariant Convolutional Networks

    Taco Cohen. “Equivariant Convolutional Networks”. Ph.D. Thesis. Available at https://pure.uva.nl/ws/files/60770359/Thesis.pdf (2021)

    work page arXiv 2021

  14. [14]

    Spherical CNNs

    Taco Cohen, Mario Geiger, Jonas Kohler, and Max Welling. “Spherical CNNs”. Proceedings of the International Conference on Learning Representations (2018)

    2018

  15. [15]

    Sheaves, Cosheaves, and Applications

    Justin Curry. “Sheaves, Cosheaves, and Applications”. Ph.D. Thesis. Arxiv Preprint (2013)

    2013

  16. [16]

    Approximation by superpositions of a sigmoidal function

    George Cybenko “Approximation by superpositions of a sigmoidal function”. Mathematics of Control, Signals, and Systems. 2 (4): 303–314 (1989)

    1989

  17. [17]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai et al. “An image is worth 16x16 words: Transformers for image recognition at scale”. Proceedings of the International Conference on Learning Representations (2021)

    2021

  18. [18]

    Adversarial Robustness of Graph Transformers

    Philipp Foth, Lukas Gosch, Simon Geisler, Leo Schwinn, and Stephan Günnemann. “Adversarial Robustness of Graph Transformers”. Transactions on Machine Learning Research (2025)

    2025

  19. [19]

    On the approximate realization of continuous mappings by neural networks

    Ken-Ichi Funahashi “On the approximate realization of continuous mappings by neural networks”. Neural Networks. 2 (3): 183–192 (1989)

    1989

  20. [20]

    Explaining and harnessing adversarial examples

    Ian J. Goodfellow, Jonathon Shlens, and Christian Szegedy “Explaining and harnessing adversarial examples”. Proceedings of the International Conference on Learning Representations (2015)

    2015

  21. [21]

    Theoretical Limitations of Self-Attention in Neural Sequence Models

    Michael Hahn. “Theoretical Limitations of Self-Attention in Neural Sequence Models”. Transactions of the Association for Computational Linguistics. Vol. 8 pp.156–171. (2020)

    2020

  22. [22]

    Sheaf Neural Networks

    Jakob Hansen and Thomas Gebhart. “Sheaf Neural Networks”. NeurIPS’20 Workshop on TDA and Beyond. (2020)

    2020

  23. [23]

    Copresheaf Topological Neural Networks: A Generalized Deep Learning Framework

    Mustafa Hajij, Lennart Bastian, Sarah Osentoski, Hardik Kabaria, John Davenport, Dawood, Balaji Cherukuri, Joseph Kocheemoolayil, Nastaran Shahmansouri, Adrian Lew, Theodore Papamarkou, and Tolga Birdal.“Copresheaf Topological Neural Networks: A Generalized Deep Learning Framework”. NeurIPS’25. Proceedings of the 39th International Conference on Neural In...

    2025

  24. [24]

    Algebraic Geometry

    Robin Hartshorne. “Algebraic Geometry”. Graduate Texts in Mathematics, Springer. 60–69 (1977)

    1977

  25. [25]

    Algebraic Topology

    Allen Hatcher. “Algebraic Topology”. Cambridge University PRess. (2002) 31

    2002

  26. [26]

    Sheaf-based Positional Encodings for Graph Neural Networks

    Yu He, Cristian Bodnar, and Pietro Liò. “Sheaf-based Positional Encodings for Graph Neural Networks”. Proceedings of the 2nd NeurIPS Workshop on Symmetry and Geometry in Neural Representations, PMLR 228:1-18. (2024)

    2024

  27. [27]

    Long short-term meory

    S. Hochreiter and J. Schmidhuber. “Long short-term meory”. Neural Computation. Vol. 9 No. 8, pp.1700-1709 (1997)

    1997

  28. [28]

    Multilayer feedforward networks are universal approxi- mators

    Kurt Hornik, Maxwell Stinchcombe, and Halbert White “Multilayer feedforward networks are universal approxi- mators”. Neural Networks. 2 (5): 359–366. (1989)

    1989

  29. [29]

    On the Robustness of Self-attentive models

    Yu-Lun Hsieh, Minhao Cheng, Da-Cheng Juan, wei Wei, Wen-Lian Hsu, and Cho-Jui Hsieh. “On the Robustness of Self-attentive models”. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. pp. 1502-001529. (2019)

    2019

  30. [30]

    Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers

    Yixiao Huang, Hanlin Zhu, Tianyu Guo, Jiantao Jiao, Somayeh Sojoudi, Michael Jordan, Stuart J Russell, and Song Mei. “Generalization or Hallucination? Understanding Out-of-Context Reasoning in Transformers”. NeurIPS’25. Proceedings of the 39th International Conference on Neural Information Processing Systems. (2025)

    2025

  31. [31]

    Fuchsian Groups

    Svetlana Katok. “Fuchsian Groups”. University of Chicago Press. pp. 63–77 (1992)

    1992

  32. [32]

    Universal Cover of a Surface (with Boundary)

    Moishe Kohan “Universal Cover of a Surface (with Boundary)”. Mathematics Stack Exchange, https://math.stackexchange.com/questions/673187/universal-cover-of-a-surface-with-boundary (version: February 12th, 2014)

    2014

  33. [33]

    Universal Approximation with Deep Narrow Networks

    Patrick Kidger and Terry Lyons. “Universal Approximation with Deep Narrow Networks”. Conference on Learning Theory. (2020)

    2020

  34. [34]

    Semi-supervised classification with graph convolutional networks

    Thomas Kipf, Max Welling “Semi-supervised classification with graph convolutional networks”. ICLR (2017)

    2017

  35. [35]

    Does there exist a continuous, open, and surjective map fromf : Rn→Rm form > n?

    Moishe Kohan. (https://math.stackexchange.com/questions/3130389/does-there-exist-a-continuous-open-and- surjective-map-from-f-colon-mathbbr). “Does there exist a continuous, open, and surjective map fromf : Rn→Rm form > n?”. Mathematics Stack Exchange. Answer provided on March 4th, 2019

    work page arXiv 2019

  36. [36]

    Universal covers, color refinement, and two-variable counting logic: Lower bounds for the depth

    Andreas Krebs and Oleg Verbitsky. “Universal covers, color refinement, and two-variable counting logic: Lower bounds for the depth". 2015 30th Annual ACM/IEEE Symposium on Logic in Computer Science. 15345831 (2015)

    2015

  37. [37]

    The Expressive Power of Neural Networks: A View from the Width

    Zhou Lu, Hongming Pu, Feicheng Wang, Zhiqiang Hu, and Liwei Wang. “The Expressive Power of Neural Networks: A View from the Width”. Advances in Neural Information Processing Systems. Vol. 30. (2017)

    2017

  38. [38]

    On the robustness of vision transformers to adversarial examples

    Kaleel Mahmood, Rigel Mahmood, and Marten van Dijk. “On the robustness of vision transformers to adversarial examples”. IEEE/CVF International Conference on Computer Vision (ICCV). (2021)

    2021

  39. [39]

    On the Vulnerability of Capsule Networks to Adversarial Attacks

    Felix Michels, Tobian Uelwer, Eric Upschulte, and Stefan Harmeling “On the Vulnerability of Capsule Networks to Adversarial Attacks”. International Conference on deep learning 2019, Workshop on Security and Privacy of deep learning (2019)

    2019

  40. [40]

    Crafting Adversarial Input Sequences for Recurrent Neural Networks

    Nicolas Papernot, Patrick McDaniel, Ananthram Swami, and Richard Harang. “Crafting Adversarial Input Sequences for Recurrent Neural Networks”. IEEE Military Communications Conference (2016)

    2016

  41. [41]

    Minimum Width for Universal Approximation

    Sejun Park, Chulhee Yun, Jaeho Lee, and Jinwoo Shin. “Minimum Width for Universal Approximation”. Proceedings of the International Conference on Learning Representations (2021)

    2021

  42. [42]

    A persistent Weisfeiler-Lehman procedure for graph classification

    Bastian Rieck, Christian Bock, Karsten Borgwardt. “A persistent Weisfeiler-Lehman procedure for graph classification”, Proceedings of the 36th International Conference on deep learning, PMLR 97:5448-5458 (2019)

    2019

  43. [43]

    Dynamic Routing between Capsules

    Sara Sabour, Nicholas Frosst, and Geoffrey E Hinton. “Dynamic Routing between Capsules”. Advances in Neural Information Processing Systems. vol.30 (2017)

    2017

  44. [44]

    Mitigating adversarial attacks on transformer models in credit scoring

    Brandon Schwab and Johannes Kriebel. “Mitigating adversarial attacks on transformer models in credit scoring”. European Journal of Operational Research 328 (1) pp. 309-323 (2026)

    2026

  45. [45]

    Weisfeiler- Lehman Graph Kernels

    Nino Shervashidze, Pascal Schweitzer, Erik Jan van Leeuwen, Kurt Mehlhorn, Karsten M. Borgwardt. “Weisfeiler- Lehman Graph Kernels”, Journal of deep learning Research no.12 (2011) 2539-2561

    2011

  46. [46]

    Cohomology of Sheaves

    Stacks Project. “Cohomology of Sheaves”. url: https://stacks.math.columbia.edu/tag/01DZ

  47. [47]

    From Noise to Narrative: Tracing the Origins of Hallucinations in Transformers

    Praneet Suresh, Jack Stanley, Sonia Joseph, Luca Scimeca, and Danilo Bzdok. “From Noise to Narrative: Tracing the Origins of Hallucinations in Transformers”. NeurIPS’25. Proceedings of the 39th International Conference on Neural Information Processing Systems. (2025)

    2025

  48. [48]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, Illia Polosukhin. “Attention is all you need”. Advances in Neural Information Processing Systems. Vol. 30. (2017)

    2017

  49. [49]

    On the limitations of representing functions on sets

    Edward Wagstaff, Fabian B. Fuchs, Martin Engelcke, Ingmar Posner, and Michael Osborne. “On the limitations of representing functions on sets”. Proceedings of the 36th International Conference on Machine Learning. (2019)

    2019

  50. [50]

    Monotone and open mappings on manifolds. I

    John Walsh. “Monotone and open mappings on manifolds. I”. Transactions of the American Mathematical Society. Vol. 209 (1975)

    1975

  51. [51]

    How Powerful are Graph Neural Networks?

    Keyulu Xu, Weihua Hu, Jure Leskovec, and Stefanie Jegelka. “How Powerful are Graph Neural Networks?”. International Conference on Learning Representations (2019). 32

    2019

  52. [52]

    Deep Sets

    Manzil Zaheer, Satwik Kottur, Siamak Ravanbhakhsh, Barnabas Poczos, Ruslan Salakhutdinov, and Alexander J Smola. “Deep Sets”. 31st Conference on Neural Information Processing Systems. (2017)

    2017

  53. [53]

    Adversarial attacks on deep learning models in natural language processing: A survey

    Wei Emma Zhang, Quan Z. Sheng, Ahoud Alhazmi, and Chenliang Li. “Adversarial attacks on deep learning models in natural language processing: A survey”. ACM Transactions on Intelligent Systems and Technology. Vol.11 No.3 pp.1-41. (2020)

    2020

  54. [54]

    Graph neural networks: A review of methods and applications

    Jie Zhou, Ganqu Cui, Shengding Hu, Zhengyan Zhang, Cheng Yang, Zhiyuan Liu, Lifeng Wang, Changcheng Li, and Maosong Sun.“Graph neural networks: A review of methods and applications”. AI Open I. 57–81 (2020)

    2020

  55. [55]

    Adversarial Attacks on Graph Neural Networks via Meta Learning

    Daniel Zugner and Stephan Gunnemann “Adversarial Attacks on Graph Neural Networks via Meta Learning”. Proceedings of the International Conference on Learning Representations (2019) Max Planck Institute for Mathematics, Vivatsgasse 7, 53111 Bonn, Germany Email address:s.park@mpim-bonn.mpg.de SolverX, Gangseo-gu, Seoul 07801, Republic of Korea Email address...

    2019