Imbalances in Neurosymbolic Learning: Characterization and Mitigating Strategies
Not on the map. Why?
Abstract (arXiv)
We study one of the most popular problems in **neurosymbolic learning** (NSL), that of learning neural classifiers given only the result of applying a symbolic component σ to the gold labels of the elements of a vector mathbf x. The gold labels of the elements in mathbf x are unknown to the learner. We make multiple contributions, theoretical and practical, to address a problem that has not been studied so far in this context, that of characterizing and mitigating *learning imbalances*, i.e., major differences in the errors that occur when classifying instances of different classes (aka **class-specific risks**). Our theoretical analysis reveals a unique phenomenon: that σ can greatly impact learning imbalances. This result sharply contrasts with previous research on supervised and weakly supervised learning, which only studies learning imbalances under data imbalances. On the practical side, we introduce a technique for estimating the marginal of the hidden gold labels using weakly supervised data. Then, we introduce algorithms that mitigate imbalances at training and testing time by treating the marginal of the hidden labels as a constraint. We demonstrate the effectiveness of our techniques using strong baselines from NSL and long-tailed learning, suggesting performance improvements of up to 14%.
Key facts. Cited by 2 papers in the map (data through 7 Oct 2026). Builds on 41 of its 76 references that are in the map. First posted on arXiv 13 Jul 2024.
Builds on
41 of 76 references.
References · 41
- 2021Convex optimizationcited by 2,479
- 2013Sinkhorn Distances: Lightspeed Computation of Optimal Transportation Distancescited by 2,277
- 2019Invariant Risk Minimizationcited by 1,832
- 2019Learning Imbalanced Datasets with Label-Distribution-Aware Margin Losscited by 987
- 2019Decoupling Representation and Classifier for Long-Tailed Recognitioncited by 806
- 2018Computational Optimal Transportcited by 765
- 2016Making Deep Neural Networks Robust to Label Noise: a Loss Correction Approachcited by 663
- 2020Long-tail learning via logit adjustmentcited by 605
- 2017A systematic study of the class imbalance problem in convolutional neural networkscited by 510
- 2019Differentiable Convex Optimization Layerscited by 458
- 2019Neural Probabilistic Logic Programming in DeepProbLogcited by 389
- 2020An Investigation of Why Overparameterization Exacerbates Spurious Correlationscited by 310
29 more
- 2020Equalization Loss for Long-Tailed Object Recognitioncited by 285
- 2017A Semantic Loss Function for Deep Learning with Symbolic Knowledgecited by 267
- 2019SATNet: Bridging deep learning and logical reasoning using a differentiable satisfiability solvercited by 173
- 2020Progressive Identification of True Labels for Partial-Label Learningcited by 102
- 2020Equalization Loss v2: A New Gradient Balance Approach for Long-tailed Object Detectioncited by 99
- 2020Provably Consistent Partial-Label Learningcited by 92
- 2023NeurASP: Embracing Neural Networks into Answer Set Programmingcited by 92
- 2017The Devil is in the Tails: Fine-grained Classification in the Wildcited by 89
- 2023Scallop: A Language for Neurosymbolic Programmingcited by 63
- 2023Not All Neuro-Symbolic Concepts Are Created Equal: Analysis and Mitigation of Reasoning Shortcutscited by 63
- 2020Closed Loop Neural-Symbolic Learning via Integrating Neural Perception, Grammar Parsing, and Symbolic Reasoningcited by 58
- 2020Neural-Symbolic Integration: A Compositional Perspectivecited by 45
- 2024BEARS Make Neuro-Symbolic Models Aware of their Reasoning Shortcutscited by 36
- 2020Structured Prediction with Partial Labelling through the Infimum Losscited by 34
- 2022SoLar: Sinkhorn Label Refinery for Imbalanced Partial-Label Learningcited by 30
- 2019Named Entity Recognition with Partially Annotated Training Datacited by 27
- 2019On the Complexity of Approximating Multimarginal Optimal Transportcited by 27
- 2021On Robust Optimal Transport: Computational Complexity and Barycenter Computationcited by 27
- 2023Long-Tailed Partial Label Learning via Dynamic Rebalancingcited by 22
- 2023Improved Logical Reasoning of Language Models via Differentiable Symbolic Programmingcited by 18
- 2024Softened Symbol Grounding for Neuro-symbolic Systemscited by 16
- 2023LASER: A Neuro-Symbolic Framework for Learning Spatial-Temporal Scene Graphs with Weak Supervisioncited by 13
- 2023On Learning Latent Models with Multi-Instance Weak Supervisioncited by 13
- 2018Backpropagating through Structured Argmax using a SPIGOTcited by 11
- 2016Estimation from Indirect Supervision with Linear Momentscited by 6
- 2020Understanding the Mechanics of SPIGOT: Surrogate Gradients for Latent Structure Learningcited by 6
- 2021Paired Examples as Indirect Supervision in Latent Decision Modelscited by 6
- 2022Local Branching Relaxation Heuristics for Integer Linear Programscited by 4
- 2023On Regularization and Inference with Label Constraintscited by 3
Built on by
Citing papers per year: 2025: 1; 2026: 1.
2 papers.