Pith. sign in

REVIEW 3 major objections 5 minor 57 references

Integrating Causality with Neurochaos Learning: Proposed Approach and Research Agenda

T0 review · 3 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This position paper argues that injecting Neurochaos Learning's chaotic GLS neurons into the COMBINE step of graph neural networks can make linked-data learning causal, improving classification, prediction, and reinforcement learning.

desk verdict A clearly written research agenda, not a demonstrated result; the abstract and conclusion overclaim, and the key transfer assumption about causality preservation from time series to graphs is unexamined. read the letter →

arxiv 2501.13763 v2 pith:KH3KTARW submitted 2025-01-23 cs.LG cs.AI

Nanjangud C. Narendra , Nithin Nagaraj This is my paper · ORCID
classification cs.LGcs.AI
keywords causalityneurochaoslearninggraphneuralnetworkslinkeddatastochasticresonanceGLSmapcausalknowledgegraphsresearchagenda
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

Deep learning's diminishing returns, its spurious correlations, and its energy cost motivate this position paper's proposal: combine causal learning with Neurochaos Learning (NL), a brain-inspired method whose chaotic GLS neurons were earlier shown to preserve Granger causality in time-series data. The paper's central idea is to bring NL into graph neural networks by replacing the COMBINE step with a 1D GLS chaotic map or injecting stochastic resonance into edge weights, so that message passing on linked data keeps causal structure intact. It argues this should improve node/graph classification, link prediction, and reinforcement learning on graphs, and it outlines a research agenda of fourteen questions needed to make the integration work. A sympathetic reader would take this as an early architectural roadmap: the payoff is concrete, but the evidence is currently a transfer of an intuition from time series to graphs.

What carries the argument

The central object is the chaos neuron: the 1D generalized Lüroth series (GLS) map, $C_{\mathrm{GLS}}(x) = x/b$ for $0 \le x < b$ and $(1-x)/(1-b)$ for $b \le x < 1$, whose chaotic trajectory fires until it matches the input stimulus and yields four features: firing time, firing rate, energy, and entropy. The paper's proposal couples this neuron to a GNN's update rule $H_v^k = \mathrm{Combine}^k(H_v^{k-1}, a_v^k)$ by making the COMBINE step a chaotic firing update, or by injecting stochastic resonance into the edge weights. The GLS map's topological transitivity—the guarantee that it will eventually match almost every stimulus—together with the earlier evidence that NL preserves Granger causality is what carries the argument that causal structure survives message passing.

What would settle it

Train a GNN whose COMBINE step is a 1D GLS chaotic map on a synthetic graph generated from a known structural causal model, and compare how well the learned embeddings or edge weights recover the ground-truth causal edges against the same architecture using a ReLU activation; if the GLS variant does not recover the causal structure at least as well, the paper's central premise is contradicted. A ready-made concrete version is the paper's own drug-drug interaction example: measure link-prediction accuracy under a shifted, confounded train/test split.

Watch

Extended reading notes

Core claim

The paper's central claim is that the causality-preserving property of Neurochaos Learning, previously demonstrated for time-series transformations, can be ported into graph neural networks to obtain better results on linked data. The route is to let chaotic neurons replace the GNN's nonlinear COMBINE activation: instead of a ReLU or sigmoid update, the aggregated neighborhood features drive a 1D GLS chaotic firing process, possibly with stochastic resonance added to edge weights, turning the graph into a stochastic one whose uncertainty follows causal paths. The paper presents this as a strawman to be investigated rather than a proven technique, and it frames the surrounding design space—causal GNNs, causal sampling, spiking causal neurons, graph neural stochastic differential equations—as open research questions for classification, prediction, and reinforcement learning.

Load-bearing premise

The load-bearing premise is that the causality-preserving behaviour of chaotic neurons, demonstrated for time-series data, survives when the same neurons are inserted into a graph neural network's message-passing step; if chaotic firing destroys or ignores causal structure during graph aggregation, the proposed integration loses its motivation.

Editorial extensions

If this is right

  • Graph node and graph-level classification could become less sensitive to spurious correlations, because chaotic firing preserves dependencies that standard nonlinear activations erase.
  • Causal sampling methods like C-GraphSAGE could be composed with NL, yielding more robust predictions with very few labeled nodes.
  • Injecting stochastic resonance into edge weights gives the GNN a principled way to express uncertainty along causal paths, effectively creating a stochastic causal graph.
  • The same integration is proposed for link prediction (for example, drug-drug interactions) and for reinforcement learning, where spiking neurons can estimate causal effects to guide reward maximization.

Reading between the lines

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

  • Beyond the paper: the most immediately testable prediction is that a GLS-COMBINE GNN outperforms a standard GCN on out-of-distribution graph benchmarks with hidden confounders; the paper does not run this test.
  • Beyond the paper: because the cited causality-preservation result is about Granger (predictive) causality in time series, the transferred property is likely observational rather than interventional or counterfactual, so the strongest gains should appear in associative prediction tasks before any claim about true causal effect identification.
  • Beyond the paper: the GLS neuron's firing time could serve as a temporal signature for when a causal link becomes active, making the proposal naturally extend to temporal knowledge graphs and streaming linked data.
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

3 major / 5 minor

Summary. This position paper proposes integrating causal learning with Neurochaos Learning (NL), specifically for graph-structured (linked) data. The authors review background on causality, structural causal models, and NL, then argue that since NL preserves Granger causality in time series (their earlier result, Ref. [11]) and since GNNs can be viewed as neural structural causal models, replacing the COMBINE function in a GNN with a chaotic map such as the 1D GLS map, or injecting stochastic resonance into edge weights, could enhance classification, prediction, and reinforcement learning on graphs. The paper presents a mindmap, a list of possible integration approaches, several illustrative examples (molecule toxicity, drug-drug interaction, RL-based drug design), and fourteen research questions grouped by topic. The conclusion claims the authors 'showed' how causality and NL can be integrated, while the body offers only a speculative research agenda with no experiments, derivations, or concrete instantiations.

Significance. If the central premise were established—that NL's causality-preservation property transfers from time series to graph-structured data through the proposed GNN modifications—this would open a genuinely novel direction at the intersection of causal machine learning, brain-inspired computing, and graph representation learning. The paper is honest about being a position paper and usefully catalogs many relevant prior results, including the authors' own peer-reviewed work on ChaosNet and NL-based causality preservation, as well as related work on causal GNNs and spiking networks. The strength lies in the clear formulation of open research questions (RQ1–RQ14). However, the central architectural claim is not supported by either formal argument or empirical evidence; the paper currently functions as a research proposal rather than a demonstration, and its significance will depend entirely on future work resolving the transfer problem identified below.

major comments (3)
  1. [Section III-C, Eq. (4) and GLS map definition] The proposed integration is underspecified at the point where it matters most. The GLS map is defined as C_GLS: [0,1) -> [0,1), a scalar map, while node representations H^v in a GNN are generally high-dimensional vectors. The paper does not state whether the GLS map is applied elementwise, to a scalar projection, or to the aggregated neighborhood feature a^v. Without this specification, the proposal cannot be implemented or tested, and the claimed connection between the chaotic map and causality preservation cannot be evaluated.
  2. [Section II-B through III-C] The load-bearing premise—that NL's causality-preservation property, demonstrated for time series in Ref. [11], transfers to graph-structured data when the GLS map replaces COMBINE—is unexamined. Granger causality is a temporal notion, and the GNN's AGGREGATE step mixes features across neighbors before COMBINE, potentially destroying the very temporal or variable-level causal structure that NL preserves in sequential inputs. The paper provides no argument or evidence that causality preservation survives message passing, and without this step the central motivation for the proposed integration collapses. This is not merely a missing experiment; it is a missing conceptual link.
  3. [Section VI] The conclusion states 'we showed how causality and NL can be integrated together,' but Section III-C presents only a list of possible approaches and research questions. No demonstration, proof, or experiment appears anywhere in the manuscript. For a position paper, such overstatement misleads readers about the state of the contribution; the claim should be rephrased to 'proposed a research agenda' or the authors should provide at least one concrete, worked instantiation of the proposed GLS-based COMBINE.
minor comments (5)
  1. [Section II-A] The sentence 'In [15]. Causality has been defined...' contains a punctuation error; the citation should be integrated into the sentence, e.g., 'In [15], causality has been defined...'.
  2. [Section III-B] The terms 'AGGREGAT Eand COM BIN E' contain formatting artifacts; they should be 'AGGREGATE and COMBINE'.
  3. [Section IV-B] The word 'Thirs' is a typo for 'Third'.
  4. [Reference [4]] The energy-consumption claim cites a blog post on Medium; a peer-reviewed source would be more appropriate for such a general claim about deep learning's energy use.
  5. [Section III-C, RQ6] RQ6 asks whether NL's causality preservation could help uncover causal probabilities, but the connection to graph-based causal discovery is left entirely open; a brief explanation of how the time-series result would apply to graphs would strengthen the question.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is a position/research-agenda paper whose premises cite independent prior empirical results; the proposed integration is an open proposal, not a prediction derived from fitted parameters.

full rationale

The paper does not derive any quantitative prediction from a fitted parameter. Its central motivating premise, that Neurochaos Learning preserves Granger causality in timeseries data, is taken from the authors' earlier peer-reviewed NeurIPS paper [11], which is an external empirical result used as evidence rather than asserted by definition. The integration proposal in Section III-C is explicitly a "non-exhaustive list of possibilities" and is framed by research questions ("How would...", "Could..."), making clear it is a research agenda, not a claimed derivation. The self-citations [9], [10], [21], and [22] support background claims about ChaosNet, stochastic resonance, and feature transformation, and they are not used to forbid alternatives or to force the conclusion. The conclusion's phrase "we showed how causality and NL can be integrated together" overstates what Section III-C actually contains, but this is a rhetorical overstatement, not a circular derivation: no equation in the paper is equivalent to another by construction, and no fitted input is renamed as a prediction. The main intellectual weakness is an unexamined transfer of causality preservation from scalar timeseries to graph message passing, but that is an evidence gap, not circularity. Therefore the circularity score is 0.

Assumptions & free parameters 0 free parameters · 4 assumptions · 0 invented entities

The central claim relies on prior empirical results rather than new free parameters. The GLS map threshold b and initial activity q from ChaosNet are inherited from prior work and are not fitted in this paper. The key axioms are domain assumptions about NL's causality preservation, GNNs as SCMs, and the classification power of chaotic neurons. No new entities are postulated.

assumptions (4)
  • domain assumption Neurochaos Learning preserves Granger causality in input timeseries data.
    Invoked in Section II-B and used as the core motivation for combining NL with causal models; taken from authors' prior NeurIPS 2022 work [11], not re-established here.
  • domain assumption Any GNN can be seen as a neural structural causal model variant.
    Stated in Section III-A.2 and attributed to [24]; the causal integration design rests on this equivalence.
  • domain assumption GLS chaotic neurons with stochastic resonance perform effective classification, even with few training samples.
    Used in Section II-B and Section III-C as the basis for replacing GNN combine functions with chaotic firing; sourced from [9], [21], and [22].
  • domain assumption Causal learning and neurochaos learning each outperform plain deep learning on their own.
    Stated in the abstract and Section I, sourced from [6], [7], [9], and [10]; not re-demonstrated in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Integrating Causality with Neurochaos Learning: Proposed Approach and Research Agenda." pith.science (2026). https://pith.science/paper/KH3KTARW

@misc{pith2026250113763,
  author       = {Pith},
  title        = {Pith review of: Integrating Causality with Neurochaos Learning: Proposed Approach and Research Agenda},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KH3KTARW}},
  note         = {Machine review of arXiv:2501.13763}
}
read the original abstract

Deep learning implemented via neural networks, has revolutionized machine learning by providing methods for complex tasks such as object detection/classification and prediction. However, architectures based on deep neural networks have started to yield diminishing returns, primarily due to their statistical nature and inability to capture causal structure in the training data. Another issue with deep learning is its high energy consumption, which is not that desirable from a sustainability perspective. Therefore, alternative approaches are being considered to address these issues, both of which are inspired by the functioning of the human brain. One approach is causal learning, which takes into account causality among the items in the dataset on which the neural network is trained. It is expected that this will help minimize the spurious correlations that are prevalent in the learned representations of deep neural networks. The other approach is Neurochaos Learning, a recent development, which draws its inspiration from the nonlinear chaotic firing intrinsic to neurons in biological neural networks (brain/central nervous system). Both approaches have shown improved results over just deep learning alone. To that end, in this position paper, we investigate how causal and neurochaos learning approaches can be integrated together to produce better results, especially in domains that contain linked data. We propose an approach for this integration to enhance classification, prediction and reinforcement learning. We also propose a set of research questions that need to be investigated in order to make this integration a reality.

Figures

Figures reproduced from arXiv: 2501.13763 by the authors.

Figure 1
Figure 1. Causality-NL Integration Mindmap [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Example of a molecule - from [32]. Classification in GNNs follows the following process. First, a GNN layer computes the input features from the input graph and generates node embeddings by aggregating the features. Node features are updated with an update function and the transformed graph is passed through a classification layer to predict labels. The most used GNN types are Graph Convolution Networks (GCN), Graph… view at source ↗
Figure 3
Figure 3. C-GraphSAGE process. C-GraphSAGE is pictorially depicted in [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Neighborhoods Sampling and Mapping to the Label Set. [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Drug-Drug Interaction example preserves causality [11] be useful here, and if so, how? IV. EXTENSIONS In this section, we discuss some extensions beyond the usual classification-based neural network-based approach. Our focus will be on predictive modeling and reinforce…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

57 extracted references · 45 canonical work pages

  1. [11]

    Causality preserving chaotic transformation and classification using neurochaos learning,

    H. NB, A. Kathpalia, and N. Nagaraj, “Causality preserving chaotic transformation and classification using neurochaos learning,” Advances in Neural Information Processing Systems , vol. 35, pp. 2046–2058, 2022

    work page 2022

  2. [1]

    Deep learning,

    Y . LeCun, Y . Bengio, and G. Hinton, “Deep learning,” nature, vol. 521, no. 7553, pp. 436–444, 2015

    2015

  3. [2]

    A survey on deep learning and its applications,

    S. Dong, P. Wang, and K. Abbas, “A survey on deep learning and its applications,” Computer Science Review , vol. 40, p. 100379, 2021

    work page 2021

  4. [3]

    Deep learning is hitting a wall,

    G. Marcus, “Deep learning is hitting a wall,” Nautilus, Accessed , pp. 03–11, 2022

    work page 2022

  5. [4]

    Ai is harming our planet; addressing ai’s staggering energy cost (2023 update) : https://medium.com/@numenta/ai-is-harming-our- planet-addressing-2023-update-6115e324d81e

    work page 2023

  6. [5]

    Causality: The next step in artificial intelligence,

    L. Cavique, “Causality: The next step in artificial intelligence,” in Philosophy of Artificial Intelligence and Its Place in Society . IGI Global, 2023, pp. 1–17

    work page 2023

  7. [6]

    Causal deep learning,

    J. Berrevoets, K. Kacprzyk, Z. Qian, and M. van der Schaar, “Causal deep learning,” arXiv preprint arXiv:2303.02186 , 2023

    arXiv 2023

  8. [7]

    When causal inference meets deep learning,

    Y . Luo, J. Peng, and J. Ma, “When causal inference meets deep learning,” Nature Machine Intelligence , vol. 2, no. 8, pp. 426–427, 2020

    work page 2020

Show all 57 references
  1. [8]

    Do humans think causally, and how?

    J. Vallverd ´u, “Do humans think causally, and how?” in Causality for Artificial Intelligence: From a Philosophical Perspective . Springer, 2024, pp. 33–42

    2024

  2. [9]

    When noise meets chaos: Stochastic resonance in neurochaos learning,

    N. B. Harikrishnan and N. Nagaraj, “When noise meets chaos: Stochastic resonance in neurochaos learning,” Neural Networks, vol. 143, pp. 425– 435, 2021

    2021

  3. [10]

    Neurochaos feature transformation for machine learning,

    D. Sethi, N. Nagaraj, and N. B. Harikrishnan, “Neurochaos feature transformation for machine learning,” Integration, vol. 90, pp. 157–162, 2023

    2023

  4. [12]

    Linked data: Principles and state of the art,

    C. Bizer, T. Heath, and T. Berners-Lee, “Linked data: Principles and state of the art,” in World wide web conference, vol. 1. Citeseer, 2008, p. 40

    2008

  5. [13]

    Opportunities for neuromorphic computing algorithms and applications,

    C. D. Schuman, S. R. Kulkarni, M. Parsa, J. P. Mitchell, P. Date, and B. Kay, “Opportunities for neuromorphic computing algorithms and applications,” Nature Computational Science , vol. 2, no. 1, pp. 10–19, 2022

    2022

  6. [14]

    Pearl and D

    J. Pearl and D. Mackenzie, The book of why: the new science of cause and effect. Basic books, 2018

    2018

  7. [15]

    Pearl, M

    J. Pearl, M. Glymour, and N. P. Jewell, Causal inference in statistics: A primer. John Wiley & Sons, 2016

    2016

  8. [16]

    An introduction to causal inference,

    J. Pearl, “An introduction to causal inference,” The international journal of biostatistics, vol. 6, no. 2, 2010

    2010

  9. [17]

    A tutorial on learning with bayesian networks,

    D. Heckerman, “A tutorial on learning with bayesian networks,” Learn- ing in graphical models , pp. 301–354, 1998

    1998

  10. [18]

    Causalkg: Causal knowledge graph explain- ability using interventional and counterfactual reasoning,

    U. Jaimini and A. Sheth, “Causalkg: Causal knowledge graph explain- ability using interventional and counterfactual reasoning,” IEEE Internet Computing, vol. 26, no. 1, pp. 43–50, 2022

    2022

  11. [19]

    Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain,

    F. A. Azevedo, L. R. Carvalho, L. T. Grinberg, J. M. Farfel, R. E. Ferretti, R. E. Leite, W. J. Filho, R. Lent, and S. Herculano-Houzel, “Equal numbers of neuronal and nonneuronal cells make the human brain an isometrically scaled-up primate brain,” Journal of Comparative Neur...

    2009

  12. [20]

    Stochastic resonance in climatic change,

    R. Benzi, G. Parisi, A. Sutera, and A. Vulpiani, “Stochastic resonance in climatic change,” Tellus, vol. 34, no. 1, pp. 10–16, 1982

    1982

  13. [21]

    Neurochaos inspired hybrid machine learning architecture for classification,

    N. Harikrishnan and N. Nagaraj, “Neurochaos inspired hybrid machine learning architecture for classification,” in 2020 International Confer- ence on Signal Processing and Communications (SPCOM) . IEEE, 2020, pp. 1–5

    2020

  14. [22]

    Chaosnet: A chaos based artificial neural network architecture for classification,

    H. N. Balakrishnan, A. Kathpalia, S. Saha, and N. Nagaraj, “Chaosnet: A chaos based artificial neural network architecture for classification,” Chaos: An Interdisciplinary Journal of Nonlinear Science , vol. 29, no. 11, 2019

    2019

  15. [23]

    What is a support vector machine?

    W. S. Noble, “What is a support vector machine?” Nature biotechnology, vol. 24, no. 12, pp. 1565–1567, 2006

    2006

  16. [24]

    Relating graph neural networks to structural causal models,

    M. Ze ˇcevi´c, D. S. Dhami, P. Veli ˇckovi´c, and K. Kersting, “Relating graph neural networks to structural causal models,” arXiv preprint arXiv:2109.04173, 2021

    2021 arXiv

  17. [25]

    Link prediction based on graph neural net- works,

    M. Zhang and Y . Chen, “Link prediction based on graph neural net- works,” Advances in neural information processing systems , vol. 31, 2018

    2018

  18. [26]

    Graph neu- ral networks and reinforcement learning: A survey,

    F. Fathinezhad, P. Adibi, B. Shoushtarian, and J. Chanussot, “Graph neu- ral networks and reinforcement learning: A survey,” in Deep Learning and Reinforcement Learning . IntechOpen, 2023

    2023

  19. [27]

    E. F. Kendall and D. L. McGuinness, Ontology engineering. Morgan & Claypool Publishers, 2019

    2019

  20. [28]

    The semantic web,

    O. Lassila, J. Hendler, and T. Berners-Lee, “The semantic web,” Scien- tific American, vol. 284, no. 5, pp. 34–43, 2001

    2001

  21. [29]

    Exploring causal learning through graph neural networks: an in-depth review,

    S. Job, X. Tao, T. Cai, H. Xie, L. Li, J. Yong, and Q. Li, “Exploring causal learning through graph neural networks: an in-depth review,” arXiv preprint arXiv:2311.14994 , 2023

    2023 arXiv

  22. [30]

    Estimation of the kullback-leibler divergence,

    M. Broniatowski, “Estimation of the kullback-leibler divergence,” Math- ematical Methods of Statistics , vol. 12, no. 4, pp. 391–409, 2003

    2003

  23. [31]

    Moleculenet dataset : https://moleculenet.org/datasets-1

  24. [32]

    Graph neural networks for graph classification: https://blog.dataiku.com/graph-neural-networks-part-three

  25. [33]

    Graph classification with graph neural networks: https : //colab.research.google.com/drive/ 1i8a0d f q3f i7njc62 mvxulcaleuclnb/

  26. [34]

    Causal graphsage: A robust graph method for classification based on causal sampling,

    T. Zhang, H.-R. Shan, and M. A. Little, “Causal graphsage: A robust graph method for classification based on causal sampling,” Pattern Recognition, vol. 128, p. 108696, 2022

    2022

  27. [35]

    Inductive representation learning on large graphs,

    W. Hamilton, Z. Ying, and J. Leskovec, “Inductive representation learning on large graphs,” Advances in neural information processing systems, vol. 30, 2017

    2017

  28. [36]

    Logistic map: A possible random-number generator,

    S. C. Phatak and S. S. Rao, “Logistic map: A possible random-number generator,” Physical review E , vol. 51, no. 4, p. 3670, 1995

    1995

  29. [37]

    Robust neural networks using stochastic resonance neurons,

    E. Manuylovich, D. Arg ¨uello Ron, M. Kamalian-Kopae, and S. K. Turitsyn, “Robust neural networks using stochastic resonance neurons,” Communications Engineering, vol. 3, no. 1, p. 169, 2024

    2024

  30. [38]

    Neural spiking for causal inference and learning,

    B. J. Lansdell and K. P. Kording, “Neural spiking for causal inference and learning,” PLOS Computational Biology, vol. 19, no. 4, p. e1011005, 2023

    2023

  31. [39]

    Signn: A spike- induced graph neural network for dynamic graph representation learn- ing,

    D. Chen, S. Zheng, M. Xu, Z. Zhu, and Y . Zhao, “Signn: A spike- induced graph neural network for dynamic graph representation learn- ing,” Pattern Recognition, vol. 158, p. 111026, 2025

    2025

  32. [40]

    Integrate-and-fire models: https://neuronaldynamics.epfl.ch/online/ch1.s3.html

  33. [41]

    Stochastic graph as a model for social networks,

    A. Rezvanian and M. R. Meybodi, “Stochastic graph as a model for social networks,” Computers in Human Behavior , vol. 64, pp. 621–640, 2016

    2016

  34. [42]

    Graph neu- ral stochastic differential equations,

    R. Bergna, F. Opolka, P. Li `o, and J. M. Hernandez-Lobato, “Graph neu- ral stochastic differential equations,” arXiv preprint arXiv:2308.12316 , 2023

    2023 arXiv

  35. [43]

    A brain-inspired causal reasoning model based on spiking neural networks,

    H. Fang and Y . Zeng, “A brain-inspired causal reasoning model based on spiking neural networks,” in 2021 International Joint Conference on Neural Networks (IJCNN) . IEEE, 2021, pp. 1–5

    2021

  36. [44]

    Random features strengthen graph neural networks,

    R. Sato, M. Yamada, and H. Kashima, “Random features strengthen graph neural networks,” in Proceedings of the 2021 SIAM international conference on data mining (SDM) . SIAM, 2021, pp. 333–341

    2021

  37. [45]

    Stochastic aggregation in graph neural networks,

    Y . Wang and T. Karaletsos, “Stochastic aggregation in graph neural networks,” arXiv preprint arXiv:2102.12648 , 2021

    2021 arXiv

  38. [46]

    Mapping the neuro- symbolic ai landscape by architectures: A handbook on augmenting deep learning through symbolic reasoning,

    J. Feldstein, P. Dilkas, V . Belle, and E. Tsamoura, “Mapping the neuro- symbolic ai landscape by architectures: A handbook on augmenting deep learning through symbolic reasoning,” arXiv preprint arXiv:2410.22077, 2024

    2024 arXiv

  39. [47]

    Emergence of scaling in random net- works,

    A.-L. Barab ´asi and R. Albert, “Emergence of scaling in random net- works,” science, vol. 286, no. 5439, pp. 509–512, 1999

    1999

  40. [48]

    Predicting missing links via local information,

    T. Zhou, L. L ¨u, and Y .-C. Zhang, “Predicting missing links via local information,” The European Physical Journal B , vol. 71, pp. 623–630, 2009

    2009

  41. [49]

    Customized subgraph selection and encoding for drug-drug interaction prediction,

    H. Du, Q. Yao, J. Zhang, Y . Liu, and Z. Wang, “Customized subgraph selection and encoding for drug-drug interaction prediction,” arXiv preprint arXiv:2411.01535, 2024

    2024 arXiv

  42. [50]

    Efficient reinforcement learning through evolving neural network topologies,

    K. O. Stanley and R. Miikkulainen, “Efficient reinforcement learning through evolving neural network topologies,” in Proceedings of the 4th Annual Conference on genetic and evolutionary computation , 2002, pp. 569–577

    2002

  43. [51]

    De novo drug design using reinforcement learning with graph-based deep generative models,

    S. R. Atance, J. V . Diez, O. Engkvist, S. Olsson, and R. Mercado, “De novo drug design using reinforcement learning with graph-based deep generative models,” Journal of chemical information and modeling, vol. 62, no. 20, pp. 4863–4872, 2022

    2022

  44. [52]

    Graph networks for molecular design,

    R. Mercado, T. Rastemo, E. Lindel ¨of, G. Klambauer, O. Engkvist, H. Chen, and E. J. Bjerrum, “Graph networks for molecular design,” Machine Learning: Science and Technology , vol. 2, no. 2, p. 025023, 2021

    2021

  45. [53]

    A purely spiking approach to reinforcement learning,

    M. Kiselev, A. Ivanitsky, and D. Larionov, “A purely spiking approach to reinforcement learning,” Cognitive Systems Research , p. 101317, 2024

    2024

  46. [54]

    Causallp: Learning causal relations with weighted knowledge graph link prediction,

    U. Jaimini, C. Henson, and A. Sheth, “Causallp: Learning causal relations with weighted knowledge graph link prediction,” arXiv preprint arXiv:2405.02327, 2024

    2024 arXiv

  47. [55]

    Relation semantic fusion in subgraph for inductive link prediction in knowledge graphs,

    H. Liu, J. Lu, T. Zhang, X. Hou, and P. An, “Relation semantic fusion in subgraph for inductive link prediction in knowledge graphs,” PeerJ Computer Science, vol. 10, p. e2324, 2024

    2024

  48. [56]

    Higher-order link prediction via light hypergraph neural network and hybrid aggregator,

    X. Rui, J. Zhuang, C. Sun, and Z. Wang, “Higher-order link prediction via light hypergraph neural network and hybrid aggregator,” Interna- tional Journal of Machine Learning and Cybernetics , pp. 1–15, 2024

    2024

  49. [57]

    Temporal knowl- edge graph completion: A survey,

    B. Cai, Y . Xiang, L. Gao, H. Zhang, Y . Li, and J. Li, “Temporal knowl- edge graph completion: A survey,” arXiv preprint arXiv:2201.08236 , 2022

    2022 arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.