pith. machine review for the scientific record. sign in

arxiv: 2605.10569 · v1 · submitted 2026-05-11 · 💻 cs.AI

Recognition: 2 theorem links

· Lean Theorem

Deep Arguing

Adam Gould , Francesca Toni
Authors on Pith no claims yet

Pith reviewed 2026-05-12 04:08 UTC · model grok-4.3

classification 💻 cs.AI
keywords neurosymbolic AIargumentation theoryinterpretable machine learningdeep learningcase-based explanationsdifferentiable reasoningclassification
0
0 comments X

The pith

Deep neural networks construct argumentation graphs in which data points support their predicted label and attack alternatives.

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

The paper introduces Deep Arguing, where neural networks build an explicit argumentation structure during classification. Each data point is assigned a label and linked to others through learned support or attack relations. Differentiable argumentation semantics allow the whole system to train end-to-end, optimizing both feature extraction and these argumentative interactions. Graph-level constraints guide the learning process to produce structures that explain predictions through concrete cases. On tabular and image datasets the method matches standard deep learning accuracy while supplying faithful case-based explanations.

Core claim

By having the network output an argumentation graph over the input examples, with edges encoding support for the correct class and attacks on wrong classes, and then applying differentiable semantics to compute the final prediction, the model learns representations and reasoning steps together. The resulting graph serves as a faithful case-based explanation because the support and attack relations directly determine the output through the semantics.

What carries the argument

The argumentation graph over data points, with support and attack edges learned from features and processed by differentiable argumentation semantics that compute the overall label assignment.

If this is right

  • The model reaches accuracy levels competitive with ordinary deep networks on both tabular and imaging classification tasks.
  • Every prediction is accompanied by an explicit graph showing which training cases support or attack the assigned label.
  • Constraints on the graph structure during training simultaneously raise predictive performance and the quality of the explanations.
  • The same end-to-end pipeline applies without modification to different data modalities.

Where Pith is reading between the lines

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

  • Inspecting the learned attack relations could surface systematic biases by revealing which groups of examples consistently undermine certain predictions.
  • The approach could be combined with existing attribution methods to cross-check whether the argumentative explanations align with gradient or perturbation-based importance scores.
  • Extending the same structure to regression or structured prediction tasks might yield case-based explanations for continuous outputs.

Load-bearing premise

That the support and attack relations the network learns actually mirror its internal decision process instead of being a separate structure that may not match how the features influence the output.

What would settle it

A controlled test in which a data point identified as strongly supportive in the argumentation graph is removed or its features altered, yet the model's prediction and confidence remain unchanged.

Figures

Figures reproduced from arXiv: 2605.10569 by Adam Gould, Francesca Toni.

Figure 1
Figure 1. Figure 1: Left: Illustrated architecture of parametrised base score and edge weight functions using a [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The subgraph of a QBAF for the CIFAR-10 dataset [ [PITH_FULL_IMAGE:figures/full_fig_p009_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Examples illustrating the minimality constraint for attacks (in [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: This figure showcases the full QBAF generated for the example in Section 7. The red [PITH_FULL_IMAGE:figures/full_fig_p018_4.png] view at source ↗
read the original abstract

Deep learning has become the dominant approach for creating high capacity, scalable models across diverse data modalities. However, because these models rely on a large number of learned parameters, tightly couple feature extraction with task objectives, and often lack explicit reasoning mechanisms, it is difficult for humans to understand how they arrive at their predictions. Understanding what representations emerge and why they arise from the training data remains an open challenge. We introduce Deep Arguing, a novel neurosymbolic approach that integrates deep learning with argumentation construction and reasoning for interpretable classification with different data modalities. In our approach deep neural networks construct an argumentation structure wherein data points support their assigned label and attack different ones. Using differentiable argumentation semantics for reasoning, the model is trained end-to-end to jointly learn feature representation and argumentative interactions. This results in argumentation structures providing faithful case-based explanations for predictions. Structure constraints over the argumentation graph guide learning, improving both interpretability and predictive performance. Experiments with tabular and imaging datasets show that Deep Arguing achieves performance competitive with standard baselines whilst offering interpretable argumentative reasoning.

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 / 1 minor

Summary. The paper introduces Deep Arguing, a neurosymbolic architecture in which a deep neural network constructs an argumentation graph over data points (arguments support their predicted label and attack alternatives). Differentiable argumentation semantics are used to reason over this graph, enabling end-to-end training that jointly optimizes feature representations and argumentative interactions. The resulting structures are claimed to deliver faithful case-based explanations while structure constraints improve both interpretability and predictive performance. Experiments on tabular and imaging datasets are reported to achieve performance competitive with standard baselines.

Significance. If the faithfulness claim holds and the imposed graph constraints demonstrably improve both accuracy and interpretability without hidden trade-offs, the work would provide a concrete bridge between high-capacity neural models and symbolic reasoning, addressing a central challenge in explainable AI. The use of differentiable semantics for joint learning of representations and interactions is a technically interesting direction that could generalize across modalities.

major comments (2)
  1. [Abstract / Experiments] Abstract and Experiments section: the central claim that the constructed argumentation structures provide 'faithful case-based explanations' is not supported by any quantitative faithfulness metric (e.g., agreement between the semantics-derived label and the network's internal activations, or ablation showing that removing the graph alters predictions in the predicted manner). Training merely aligns the auxiliary structure to the network output; without an independent verification test, it remains possible that the graph is an imposed constraint whose output is post-hoc aligned rather than a faithful reflection of the model's reasoning.
  2. [Experiments] Experiments section: performance is described only as 'competitive with standard baselines' with no tables, numerical results, error bars, specific baselines, or statistical tests. This prevents assessment of whether the structure constraints deliver the claimed performance improvement or merely preserve accuracy while adding constraints, directly undermining the dual claim of improved interpretability and predictive performance.
minor comments (1)
  1. [Abstract] The abstract states that 'structure constraints over the argumentation graph guide learning' but does not specify the exact form of these constraints or how they are enforced during back-propagation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on our manuscript. The comments highlight important areas for strengthening the presentation of faithfulness claims and experimental details. We address each major comment below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: [Abstract / Experiments] Abstract and Experiments section: the central claim that the constructed argumentation structures provide 'faithful case-based explanations' is not supported by any quantitative faithfulness metric (e.g., agreement between the semantics-derived label and the network's internal activations, or ablation showing that removing the graph alters predictions in the predicted manner). Training merely aligns the auxiliary structure to the network output; without an independent verification test, it remains possible that the graph is an imposed constraint whose output is post-hoc aligned rather than a faithful reflection of the model's reasoning.

    Authors: We thank the referee for this observation. In Deep Arguing, the argumentation graph is not an auxiliary or post-hoc construct: the neural network produces the graph, and the final classification is computed directly via differentiable argumentation semantics over the support and attack relations. The explanations are therefore faithful by construction, as the predicted label is a direct function of the learned argumentative interactions rather than an independent alignment step. The joint end-to-end training under structure constraints further ensures that the graph reflects the model's reasoning. To provide quantitative support as requested, we will add an ablation study (performance with vs. without the argumentation module) and a faithfulness metric (e.g., agreement rate between semantics-derived labels and direct network outputs) to the Experiments section. revision: yes

  2. Referee: [Experiments] Experiments section: performance is described only as 'competitive with standard baselines' with no tables, numerical results, error bars, specific baselines, or statistical tests. This prevents assessment of whether the structure constraints deliver the claimed performance improvement or merely preserve accuracy while adding constraints, directly undermining the dual claim of improved interpretability and predictive performance.

    Authors: We acknowledge that the current version does not present experimental results with sufficient detail. Although the manuscript includes experiments on tabular and imaging datasets, we agree that the lack of explicit tables, numerical values, error bars, named baselines, and statistical tests limits evaluation of the claimed benefits. In the revised manuscript we will expand the Experiments section with comprehensive tables reporting accuracies (with standard deviations), comparisons against specific baselines (e.g., standard DNNs and other neurosymbolic methods), error bars across multiple runs, and statistical significance tests. This will allow direct assessment of whether the structure constraints improve or maintain predictive performance while enhancing interpretability. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained via explicit neurosymbolic design

full rationale

The paper defines a method in which a neural network explicitly constructs an argumentation graph (data points as arguments) and applies differentiable semantics to produce the label prediction, with end-to-end training aligning the two. The claim that the resulting structures provide 'faithful case-based explanations' follows directly from this construction rather than from any hidden reduction or self-referential fit. No equations, self-citations, or uniqueness theorems are invoked in the abstract or described text that would make a central result equivalent to its inputs by definition. The approach is presented as an imposed auxiliary structure whose outputs are aligned by training, which is a standard design choice rather than circularity. External benchmarks (competitive performance on tabular/imaging data) are referenced without reducing to internal parameters.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

The approach rests on the assumption that differentiable argumentation semantics can be defined and that the resulting structures remain faithful explanations; no free parameters or invented entities are quantified in the abstract.

axioms (1)
  • domain assumption Differentiable versions of standard argumentation semantics exist and preserve the intended support/attack semantics during gradient-based training.
    Invoked to enable end-to-end training of the joint feature and argument model.
invented entities (1)
  • Argumentation structure constructed by the DNN no independent evidence
    purpose: To encode support and attack relations that yield faithful case-based explanations.
    Introduced as the core output of the network; no independent evidence of faithfulness is supplied in the abstract.

pith-pipeline@v0.9.0 · 5464 in / 1286 out tokens · 28425 ms · 2026-05-12T04:08:29.520429+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

61 extracted references · 61 canonical work pages

  1. [1]

    Optuna: A next-generation hyperparameter optimization framework

    Takuya Akiba, Shotaro Sano, Toshihiko Yanase, Takeru Ohta, and Masanori Koyama. Optuna: A next-generation hyperparameter optimization framework. InProceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’19, page 2623–2631, New York, NY , USA, 2019. Association for Computing Machinery

    work page 2019

  2. [2]

    Gradual Semantics Accounting for Varied-Strength Attacks

    Leila Amgoud and Dragan Doder. Gradual Semantics Accounting for Varied-Strength Attacks. In18th International Conference on Autonomous Agents and MultiAgent Systems (AAMAS 2019), pages 1270–1278, Montréal, Canada, May 2019. IFAAMAS : International Foundation for Autonomous Agents and Multiagent Systems and SIGAI : ACM’s Special Interest Group on Artificia...

    work page 2019

  3. [3]

    Protoargnet: Interpretable image classification with super-prototypes and argumentation [technical report], 2023

    Hamed Ayoobi, Nico Potyka, and Francesca Toni. Protoargnet: Interpretable image classification with super-prototypes and argumentation [technical report], 2023

    work page 2023

  4. [4]

    Sparx: Sparse argumentative explanations for neural networks

    Hamed Ayoobi, Nico Potyka, and Francesca Toni. Sparx: Sparse argumentative explanations for neural networks. InECAI 2023, pages 149–156. IOS Press, 2023

    work page 2023

  5. [5]

    Logic tensor networks.Artificial Intelligence, 303:103649, February 2022

    Samy Badreddine, Artur d’Avila Garcez, Luciano Serafini, and Michael Spranger. Logic tensor networks.Artificial Intelligence, 303:103649, February 2022

    work page 2022

  6. [6]

    Pietro Baroni, Antonio Rago, and Francesca Toni. How many properties do we need for gradual argumentation? InProceedings of the Thirty-Second AAAI Conference on Artificial Intelligence and Thirtieth Innovative Applications of Artificial Intelligence Conference and Eighth AAAI Symposium on Educational Advances in Artificial Intelligence, AAAI’18/IAAI’18/EA...

    work page 2018

  7. [7]

    Adult, 1996

    Ronny Kohavi Barry Becker. Adult, 1996

    work page 1996

  8. [8]

    Covertype, 1998

    Jock Blackard. Covertype, 1998

    work page 1998

  9. [9]

    Data-empowered argumentation for dialectically explainable predictions

    Oana Cocarascu, Andria Stylianou, Kristijonas ˇCyras, and Francesca Toni. Data-empowered argumentation for dialectically explainable predictions. InECAI 2020, pages 2449–2456. IOS Press, 2020

    work page 2020

  10. [10]

    Explanatory predictions with artificial neural networks and argumentation

    Oana Cocarascu, Kristijonas ˇCyras, and Francesca Toni. Explanatory predictions with artificial neural networks and argumentation. InProceedings of the 2ndWorkshop onExplainable Artificial Intelligence (XAI 2018), May 2018

    work page 2018

  11. [11]

    Neuro-symbolic learning of answer set programs from raw data

    Daniel Cunnington, Mark Law, Jorge Lobo, and Alessandra Russo. Neuro-symbolic learning of answer set programs from raw data. InProceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI ’23, 2023

    work page 2023

  12. [12]

    Deep symbolic learning: Discovering symbols and rules from perceptions

    Alessandro Daniele, Tommaso Campari, Sagar Malhotra, and Luciano Serafini. Deep symbolic learning: Discovering symbols and rules from perceptions. InProceedings of the Thirty- Second International Joint Conference on Artificial Intelligence, IJCAI-2023, page 3597–3605. International Joint Conferences on Artificial Intelligence Organization, August 2023

    work page 2023

  13. [13]

    d’Avila Garcez, Dov M

    Artur S. d’Avila Garcez, Dov M. Gabbay, and Luis C. Lamb. Value-based argumentation frame- works as neural-symbolic learning systems.Journal of Logic and Computation, 15(6):1041– 1058, 2005

    work page 2005

  14. [14]

    d’Avila Garcez, Dov M

    Artur S. d’Avila Garcez, Dov M. Gabbay, and Luis C. Lamb. A neural cognitive model of argumentation with application to legal inference and decision making.Journal of Applied Logic, 12(2):109–127, June 2014. 10

    work page 2014

  15. [15]

    Object-centric case- based reasoning via argumentation

    Gabriel de Olim Gaul, Adam Gould, Avinash Kori, and Francesca Toni. Object-centric case- based reasoning via argumentation. In Timotheus Kampik, Antonio Rago, Kristijonas Cyras, and Oana Cocarascu, editors,Proceedings of the 3rd International Workshop on Argumentation for eXplainable AI (ArgXAI 2025) co-located with the 28th European Conference on Artific...

    work page 2025

  16. [16]

    On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games.Artificial Intelligence, 77(2):321–357, 1995

    Phan Minh Dung. On the acceptability of arguments and its fundamental role in nonmonotonic reasoning, logic programming and n-person games.Artificial Intelligence, 77(2):321–357, 1995

    work page 1995

  17. [17]

    Glioma grading clinical and mutation features, 2022

    Kevin Camphausen Erdal Tasci. Glioma grading clinical and mutation features, 2022

    work page 2022

  18. [18]

    On interpretability of artificial neural networks: A survey, 2020

    Fenglei Fan, Jinjun Xiong, Mengzhou Li, and Ge Wang. On interpretability of artificial neural networks: A survey, 2020

    work page 2020

  19. [19]

    Argumentative large language models for explainable and contestable decision-making, 2024

    Gabriel Freedman, Adam Dejl, Deniz Gorur, Xiang Yin, Antonio Rago, and Francesca Toni. Argumentative large language models for explainable and contestable decision-making, 2024

    work page 2024

  20. [20]

    Artur d’Avila Garcez and Luís C. Lamb. Neurosymbolic ai: the 3rd wave.Artificial Intelligence Review, 56(11):12387–12406, March 2023

    work page 2023

  21. [21]

    right to explanation

    Bryce Goodman and Seth Flaxman. European union regulations on algorithmic decision making and a “right to explanation”.AI Magazine, 38(3):50–57, September 2017

    work page 2017

  22. [22]

    Preference-Based Abstract Argumentation for Case-Based Reasoning

    Adam Gould, Guilherme Paulino-Passos, Seema Dadhania, Matthew Williams, and Francesca Toni. Preference-Based Abstract Argumentation for Case-Based Reasoning. InKR, pages 394–404, 8 2024

    work page 2024

  23. [23]

    Neuro-argumentative learning with case-based reasoning

    Adam Gould and Francesca Toni. Neuro-argumentative learning with case-based reasoning. In Leilani H. Gilpin, Eleonora Giunchiglia, Pascal Hitzler, and Emile van Krieken, editors, Proceedings of The 19th International Conference on Neurosymbolic Learning and Reasoning, volume 284 ofProceedings of Machine Learning Research, pages 1090–1106. PMLR, 08–10 Sep 2025

    work page 2025

  24. [24]

    Deep residual learning for im- age recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for im- age recognition. InProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), June 2016

    work page 2016

  25. [25]

    Kipf and Max Welling

    Thomas N. Kipf and Max Welling. Semi-supervised classification with graph convolutional networks. InInternational Conference on Learning Representations, 2017

    work page 2017

  26. [26]

    Learning multiple layers of features from tiny images

    Alex Krizhevsky. Learning multiple layers of features from tiny images. Technical Report TR-2009, University of Toronto, Computer Science Department, 2009

    work page 2009

  27. [27]

    LeCun, L

    Y . LeCun, L. Bottou, Y . Bengio, and P. Haffner. Gradient-based learning applied to document recognition.Proceedings of the IEEE, 86(11):2278–2324, 1998

    work page 1998

  28. [28]

    The mnist database of handwritten digits.http://yann

    Yann LeCun. The mnist database of handwritten digits.http://yann. lecun. com/exdb/mnist/

    work page

  29. [29]

    $\mathcal{M}^4$: A unified XAI benchmark for faithfulness evaluation of feature attribution methods across metrics, modalities and models

    Xuhong Li, Mengnan Du, Jiamin Chen, Yekun Chai, Himabindu Lakkaraju, and Haoyi Xiong. $\mathcal{M}^4$: A unified XAI benchmark for faithfulness evaluation of feature attribution methods across metrics, modalities and models. InThirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track, 2023

    work page 2023

  30. [30]

    Least squares quantization in pcm.IEEE transactions on information theory, 28(2):129–137, 1982

    Stuart Lloyd. Least squares quantization in pcm.IEEE transactions on information theory, 28(2):129–137, 1982

    work page 1982

  31. [31]

    Decoupled weight decay regularization

    Ilya Loshchilov and Frank Hutter. Decoupled weight decay regularization. InInternational Conference on Learning Representations, 2019

    work page 2019

  32. [32]

    Multivariate observations

    J MacQueen. Multivariate observations. InProceedings ofthe 5th Berkeley symposium on mathematical statisticsand probability, volume 1, pages 281–297. University of California press Oakland, CA, USA, 1967. 11

    work page 1967

  33. [33]

    Neural probabilistic logic programming in deepproblog.Artificial Intelligence, 298:103504, 2021

    Robin Manhaeve, Sebastijan Duman ˇci´c, Angelika Kimmig, Thomas Demeester, and Luc De Raedt. Neural probabilistic logic programming in deepproblog.Artificial Intelligence, 298:103504, 2021

    work page 2021

  34. [34]

    Not all neuro- symbolic concepts are created equal: Analysis and mitigation of reasoning shortcuts

    Emanuele Marconato, Stefano Teso, Antonio Vergari, and Andrea Passerini. Not all neuro- symbolic concepts are created equal: Analysis and mitigation of reasoning shortcuts. In Thirty-seventh Conference on Neural Information Processing Systems, 2023

    work page 2023

  35. [35]

    Chess (king-rook vs

    Arthur Hoff Michael Bain. Chess (king-rook vs. king), 1994

    work page 1994

  36. [36]

    Guaranteed bounds on information-theoretic measures of univariate mixtures using piecewise log-sum-exp inequalities.Entropy, 18(12):442, December 2016

    Frank Nielsen and Ke Sun. Guaranteed bounds on information-theoretic measures of univariate mixtures using piecewise log-sum-exp inequalities.Entropy, 18(12):442, December 2016

    work page 2016

  37. [37]

    Curran Associates Inc., Red Hook, NY , USA, 2019

    Adam Paszke, Sam Gross, Francisco Massa, Adam Lerer, James Bradbury, Gregory Chanan, Trevor Killeen, Zeming Lin, Natalia Gimelshein, Luca Antiga, Alban Desmaison, Andreas Köpf, Edward Yang, Zach DeVito, Martin Raison, Alykhan Tejani, Sasank Chilamkurthy, Benoit Steiner, Lu Fang, Junjie Bai, and Soumith Chintala.PyTorch: an imperative style, high-performan...

    work page 2019

  38. [38]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M. Brucher, M. Perrot, and E. Duchesnay. Scikit-learn: Machine learning in Python.Journal of Machine Learning Research, 12:2825–2830, 2011

    work page 2011

  39. [39]

    Deep differentiable logic gate networks

    Felix Petersen, Christian Borgelt, Hilde Kuehne, and Oliver Deussen. Deep differentiable logic gate networks. In Alice H. Oh, Alekh Agarwal, Danielle Belgrave, and Kyunghyun Cho, editors, Advances in Neural Information Processing Systems, 2022

    work page 2022

  40. [40]

    Continuous dynamical systems for weighted bipolar argumentation

    Nico Potyka. Continuous dynamical systems for weighted bipolar argumentation. In Michael Thielscher, Francesca Toni, and Frank Wolter, editors,Principles of Knowledge Representation and Reasoning: Proceedings of the Sixteenth International Conference, KR 2018, Tempe, Arizona, 30 October - 2 November 2018, pages 148–157. AAAI Press, 2018

    work page 2018

  41. [41]

    Extending modular semantics for bipolar weighted argumentation

    Nico Potyka. Extending modular semantics for bipolar weighted argumentation. InProceedings of the 18th International Conference on Autonomous Agents and MultiAgent Systems, AAMAS ’19, page 1722–1730, Richland, SC, 2019. International Foundation for Autonomous Agents and Multiagent Systems

    work page 2019

  42. [42]

    Interpreting neural networks as quantitative argumentation frameworks.Proceed- ings of the AAAI Conference on Artificial Intelligence, 35(7):6463–6470, May 2021

    Nico Potyka. Interpreting neural networks as quantitative argumentation frameworks.Proceed- ings of the AAAI Conference on Artificial Intelligence, 35(7):6463–6470, May 2021

    work page 2021

  43. [43]

    A roadmap for neuro-argumentative learning

    Maurizio Proietti and Francesca Toni. A roadmap for neuro-argumentative learning. InNeSy, pages 1–8, 2023

    work page 2023

  44. [44]

    P. Rita S. Moro. Bank marketing, 2014

    work page 2014

  45. [45]

    The graph neural network model.IEEE Transactions on Neural Networks, 20(1):61–80, 2009

    Franco Scarselli, Marco Gori, Ah Chung Tsoi, Markus Hagenbuchner, and Gabriele Monfardini. The graph neural network model.IEEE Transactions on Neural Networks, 20(1):61–80, 2009

    work page 2009

  46. [46]

    Analyzing differentiable fuzzy logic operators.Artificial Intelligence, 302:103602, January 2022

    Emile van Krieken, Erman Acar, and Frank van Harmelen. Analyzing differentiable fuzzy logic operators.Artificial Intelligence, 302:103602, January 2022

    work page 2022

  47. [47]

    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

    work page 2018

  48. [48]

    Towards data-and knowledge-driven ai: A survey on neuro-symbolic computing.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(2):878–899, February 2025

    Wenguan Wang, Yi Yang, and Fei Wu. Towards data-and knowledge-driven ai: A survey on neuro-symbolic computing.IEEE Transactions on Pattern Analysis and Machine Intelligence, 47(2):878–899, February 2025

    work page 2025

  49. [49]

    Transparent classification with multilayer logical perceptrons and random binarization, 2019

    Zhuo Wang, Wei Zhang, Ning Liu, and Jianyong Wang. Transparent classification with multilayer logical perceptrons and random binarization, 2019. 12

    work page 2019

  50. [50]

    Scalable rule-based representation learning for interpretable classification

    Zhuo Wang, Wei Zhang, Ning Liu, and Jianyong Wang. Scalable rule-based representation learning for interpretable classification. In A. Beygelzimer, Y . Dauphin, P. Liang, and J. Wortman Vaughan, editors,Advances in Neural Information Processing Systems, 2021

    work page 2021

  51. [51]

    Tree-structured parzen estimator: Understanding its algorithm components and their roles for better empirical performance, 2023

    Shuhei Watanabe. Tree-structured parzen estimator: Understanding its algorithm components and their roles for better empirical performance, 2023

    work page 2023

  52. [52]

    Higgs, 2014

    Daniel Whiteson. Higgs, 2014

    work page 2014

  53. [53]

    Tom Nuno Wolf, Fabian Bongratz, Anne-Marie Rickmann, Sebastian Pölsterl, and Christian Wachinger. Keep the faith: Faithful explanations in convolutional neural networks for case-based reasoning.Proceedings of the AAAI Conference on Artificial Intelligence, 38(6):5921–5929, March 2024

    work page 2024

  54. [54]

    Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms, 2017

    Han Xiao, Kashif Rasul, and Roland V ollgraf. Fashion-mnist: a novel image dataset for benchmarking machine learning algorithms, 2017

    work page 2017

  55. [55]

    A semantic loss function for deep learning with symbolic knowledge

    Jingyi Xu, Zilu Zhang, Tal Friedman, Yitao Liang, and Guy Van den Broeck. A semantic loss function for deep learning with symbolic knowledge. In Jennifer Dy and Andreas Krause, editors,Proceedings of the 35th International Conference on Machine Learning, volume 80 of Proceedings of Machine Learning Research, pages 5502–5511. PMLR, 10–15 Jul 2018

    work page 2018

  56. [56]

    A survey on neural network interpretability

    Yu Zhang, Peter Tiˇno, Aleš Leonardis, and Ke Tang. A survey on neural network interpretability. IEEE Transactions on Emerging Topics in Computational Intelligence, 5(5):726–742, 2021

    work page 2021

  57. [57]

    Dags with no tears: Contin- uous optimization for structure learning

    Xun Zheng, Bryon Aragam, Pradeep K Ravikumar, and Eric Xing. Dags with no tears: Contin- uous optimization for structure learning. In S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, editors,Advances in Neural Information Processing Systems, volume 31. Curran Associates, Inc., 2018

    work page 2018

  58. [58]

    A survey on graph structure learning: Progress and opportunities, 2021

    Yanqiao Zhu, Weizhi Xu, Jinghao Zhang, Yuanqi Du, Jieyu Zhang, Qiang Liu, Carl Yang, and Shu Wu. A survey on graph structure learning: Progress and opportunities, 2021

    work page 2021

  59. [59]

    Ar- gumentative xai: A survey

    Kristijonas ˇCyras, Antonio Rago, Emanuele Albini, Pietro Baroni, and Francesca Toni. Ar- gumentative xai: A survey. InProceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-2021. International Joint Conferences on Artificial Intelligence Organization, August 2021

    work page 2021

  60. [60]

    Abstract argumentation for case-based reasoning

    Kristijonas ˇCyras, Ken Satoh, and Francesca Toni. Abstract argumentation for case-based reasoning. InFifteenth International Conference on the Principles of Knowledge Representation and Reasoning. AAAI Press, 2016. A Minimality condition The minimality condition is described in Section 4.1. Here we provide an illustration of the scenarios where minimalit...

    work page 2016

  61. [61]

    The output layer for the edge weight function had 64 neurons

    Then the base score and edge weight functions are made up of a four layer MLP with the first hidden layer having 64 neurons, the second hidden layer 48 neurons and the third hidden layer 32 neurons. The output layer for the edge weight function had 64 neurons. The ResNet was pre-trained on the CIFAR-10 data using a linear layer at the end that was removed...

    work page