mlsanityData to 7 Oct 2026

Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcuts

arXiv 2305.19951 ↗First posted 31 May 2023cs.LG stat.MLcited by 63

Not on the map. Why?

Abstract (arXiv)

Neuro-Symbolic (NeSy) predictive models hold the promise of improved compliance with given constraints, systematic generalization, and interpretability, as they allow to infer labels that are consistent with some prior knowledge by reasoning over high-level concepts extracted from sub-symbolic inputs. It was recently shown that NeSy predictors are affected by reasoning shortcuts: they can attain high accuracy but by leveraging concepts with unintended semantics, thus coming short of their promised advantages. Yet, a systematic characterization of reasoning shortcuts and of potential mitigation strategies is missing. This work fills this gap by characterizing them as unintended optima of the learning objective and identifying four key conditions behind their occurrence. Based on this, we derive several natural mitigation strategies, and analyze their efficacy both theoretically and empirically. Our analysis shows reasoning shortcuts are difficult to deal with, casting doubts on the trustworthiness and interpretability of existing NeSy solutions.

Key facts. Cited by 63 papers in the map (data through 7 Oct 2026). Builds on 60 of its 94 references that are in the map. First posted on arXiv 31 May 2023.

Builds on

60 of 94 references.

References · 60

  1. 2014Adam: A Method for Stochastic Optimizationcited by 36,960
  2. 2019PyTorch: An Imperative Style, High-Performance Deep Learning Librarycited by 20,160
  3. 2015Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networkscited by 9,617
  4. 2020Shortcut Learning in Deep Neural Networkscited by 1,928
  5. 2020Concept Bottleneck Modelscited by 1,022
  6. 2019When Does Label Smoothing Help?cited by 985
  7. 2018Challenging Common Assumptions in the Unsupervised Learning of Disentangled Representationscited by 929
  8. 2018This Looks Like That: Deep Learning for Interpretable Image Recognitioncited by 731
  9. 2019Variational Autoencoders and Nonlinear ICA: A Unifying Frameworkcited by 494
  10. 2019Neural Probabilistic Logic Programming in DeepProbLogcited by 389
  11. 2019Unmasking Clever Hans Predictors and Assessing What Machines Really Learncited by 359
  12. 2020Calibrating Deep Neural Networks using Focal Losscited by 353
38 more
  1. 2020Noise or Signal: The Role of Image Backgrounds in Object Recognitioncited by 342
  2. 2021Self-Supervised Learning with Data Augmentations Provably Isolates Content from Stylecited by 308
  3. 2017Right for the Right Reasons: Training Differentiable Models by Constraining their Explanationscited by 293
  4. 2017A Semantic Loss Function for Deep Learning with Symbolic Knowledgecited by 267
  5. 2022Mitigating Neural Network Overconfidence with Logit Normalizationcited by 251
  6. 2020Weakly-Supervised Disentanglement Without Compromisescited by 237
  7. 2020Concept Whitening for Interpretable Image Recognitioncited by 225
  8. 2017Neural-Symbolic Learning and Reasoning: A Survey and Interpretationcited by 195
  9. 2019Self-Driving Cars: A Surveycited by 147
  10. 2020Making deep neural networks right for the right scientific reasons by interacting with their explanationscited by 141
  11. 2022Interventional Causal Representation Learningcited by 140
  12. 2020Learning explanations that are hard to varycited by 132
  13. 2019From Variational to Deterministic Autoencoderscited by 131
  14. 2018Robustly Disentangled Causal Mechanisms: Validating Deep Representations for Interventional Robustnesscited by 108
  15. 2022Semantic Probabilistic Layers for Neuro-Symbolic Learningcited by 103
  16. 2021Independent mechanism analysis, a new concept?cited by 96
  17. 2017Logic Tensor Networks for Semantic Image Interpretationcited by 92
  18. 2020Explainable Object-induced Action Decision for Autonomous Vehiclescited by 88
  19. 2019Weakly Supervised Disentanglement with Guaranteescited by 82
  20. 2020On Mixup Regularizationcited by 82
  21. 2022Concept Bottleneck Model with Additional Unsupervised Conceptscited by 79
  22. 2022GlanceNets: Interpretabile, Leak-proof Concept-based Modelscited by 73
  23. 2020Coherent Hierarchical Multi-Label Classification Networkscited by 65
  24. 2022Concept Embedding Models: Beyond the Accuracy-Explainability Trade-Offcited by 61
  25. 2020Closed Loop Neural-Symbolic Learning via Integrating Neural Perception, Grammar Parsing, and Symbolic Reasoningcited by 58
  26. 2021DeepStochLog: Neural Stochastic Logic Programmingcited by 58
  27. 2021Incorporating Domain Knowledge into Deep Neural Networkscited by 56
  28. 2023Causal Component Analysiscited by 56
  29. 2022Function Classes for Identifiable Nonlinear Independent Component Analysiscited by 53
  30. 2022Leveraging Explanations in Interactive Machine Learning: An Overviewcited by 52
  31. 2022A-NeSI: A Scalable Approximate Method for Probabilistic Neurosymbolic Inferencecited by 50
  32. 2023Nonlinear Independent Component Analysis for Principled Disentanglement in Unsupervised Deep Learningcited by 48
  33. 2021MultiplexNet: Towards Fully Satisfied Logical Constraints in Neural Networkscited by 44
  34. 2022Deep Learning with Logical Constraintscited by 44
  35. 2022Synergies between Disentanglement and Sparsity: Generalization and Identifiability in Multi-Task Learningcited by 44
  36. 2020Semi-Supervised StyleGAN for Disentanglement Learningcited by 43
  37. 2022NeuPSL: Neural Probabilistic Soft Logiccited by 34
  38. 2023Neuro-Symbolic Continual Learning: Knowledge, Reasoning Shortcuts and Concept Rehearsalcited by 33

50 most cited of 60 All 60 (JSON)

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2023: 2; 2024: 14; 2025: 23; 2026: 24.

63 papers.

Sort:

By area:

  1. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMscited by 246
  2. 2024BEARS Make Neuro-Symbolic Models Aware of their Reasoning Shortcutscited by 36
  3. 2025Neuro-Symbolic AI in 2024: A Systematic Reviewcited by 36
  4. 2024Logically Consistent Language Models via Neuro-Symbolic Integrationcited by 30
  5. 2023Interpretability is in the Mind of the Beholder: A Causal Framework for Human-interpretable Representation Learningcited by 22
  6. 2024What is the Relationship between Tensor Factorizations and Circuits (and How Can We Exploit it)?cited by 22
  7. 2024Sum of Squares Circuitscited by 21
  8. 2024A Neuro-Symbolic Benchmark Suite for Concept Quality and Reasoning Shortcutscited by 19
  9. 2024On the Independence Assumption in Neurosymbolic Learningcited by 18
  10. 2025If Concept Bottlenecks are the Question, are Foundation Models the Answer?cited by 15
  11. 2025Shortcuts and Identifiability in Concept-based Models from a Neuro-Symbolic Lenscited by 13
  12. 2025Neurosymbolic Diffusion Modelscited by 12
38 more
  1. 2024The Role of Foundation Models in Neuro-Symbolic Learning and Reasoningcited by 11
  2. 2025A Probabilistic Neuro-symbolic Layer for Algebraic Constraint Satisfactioncited by 11
  3. 2024Do Concept Bottleneck Models Respect Localities?cited by 10
  4. 2025Object Centric Concept Bottleneckscited by 10
  5. 2024DeiSAM: Segment Anything with Deictic Promptingcited by 9
  6. 2023Deciphering Raw Data in Neuro-Symbolic Learning with Provable Guaranteescited by 6
  7. 2025Neural Probabilistic Circuits: Enabling Compositional and Interpretable Predictions through Logical Reasoningcited by 6
  8. 2024Understanding the Logic of Direct Preference Alignment through Logiccited by 5
  9. 2025Do ImageNet-trained models learn shortcuts? The impact of frequency shortcuts on generalizationcited by 5
  10. 2025How to Square Tensor Networks and Circuits Without Squaring Themcited by 5
  11. 2025On the Promise for Assurance of Differentiable Neurosymbolic Reasoning Paradigmscited by 3
  12. 2025A Learnability Analysis on Neuro-Symbolic Learningcited by 3
  13. 2025The Road to Generalizable Neuro-Symbolic Learning Should be Paved with Foundation Modelscited by 3
  14. 2025Interpretability as Alignment: Making Internal Understanding a Design Principlecited by 3
  15. 2025Nonparametric Identification of Latent Conceptscited by 3
  16. 2025Right for the Right Reasons: Avoiding Reasoning Shortcuts via Prototypical Neurosymbolic AIcited by 3
  17. 2026GNN Explanations that do not Explain and How to find Themcited by 3
  18. 2024Imbalances in Neurosymbolic Learning: Characterization and Mitigating Strategiescited by 2
  19. 2024MARS: A neurosymbolic approach for interpretable drug discoverycited by 2
  20. 2025Towards Reasonable Concept Bottleneck Modelscited by 2
  21. 2025Neurosymbolic Object-Centric Learning with Distant Supervisioncited by 2
  22. 2025Neuro-Symbolic AI for Cybersecurity: State of the Art, Challenges, and Opportunitiescited by 2
  23. 2026To Neuro-Symbolic Classification and Beyond by Compiling Description Logic Ontologies to Probabilistic Circuitscited by 2
  24. 2024A Neurosymbolic Approach to Adaptive Feature Extraction in SLAMcited by 1
  25. 2025Curriculum Abductive Learningcited by 1
  26. 2026The Scaling Properties of Implicit Deductive Reasoning in Transformerscited by 1
  27. 2026From Token Probabilities to Semantic Constraints: Towards Declarative Probabilistic Evaluation of Language Modelscited by 1
  28. 2024Learning Predictive Checklists with Probabilistic Logic Programmingcited by 0
  29. 2025HorNets: Learning from Discrete and Continuous Signals with Routing Neural Networkscited by 0
  30. 2025Neuro-symbolic Weak Supervision: Theory and Semanticscited by 0
  31. 2025Neural Concept Verifier: Scaling Prover-Verifier Games via Concept Encodingscited by 0
  32. 2026On Improving Neurosymbolic Learning by Exploiting the Representation Spacecited by 0
  33. 2026Logit Distance Bounds Representational Similaritycited by 0
  34. 2026Can VLMs Reason Robustly? A Neuro-Symbolic Investigationcited by 0
  35. 2026Position: Explainable AI is Causality in Disguisecited by 0
  36. 2026Constraint-Based Analysis of Reasoning Shortcuts in Neurosymbolic Learningcited by 0
  37. 2026Deep Arguingcited by 0
  38. 2026Concise and Logically Consistent Conformal Sets for Neuro-Symbolic Concept-Based Modelscited by 0

50 most cited of 63 All 63 (JSON)