Resources
Here, we include some useful resources for reasearch in concept representation learning.
Relevant Papers and Materials
Below we include a list of works in Concept representation learning, particularly in the areas of Interpretability/Explainability, that are relevant to concept-based interpretable deep learning. We will discuss several of these papers in our tutorial, however we thought that it may be benefitial to write them down in list format for people to access these works more easily. Please keep in mind that this is in no way an exhaustive list of important works within concept learning as this is a fast moving field and we have only so much space we can use here. Nevertheless, we still hope you may find this list helpful if you want to get a sense of where the field is and where it is heading.
Surveys
These are some of the surveys that touch on concept representation learning and its use in interpretable/explainable AI:
2023
- Concept-based Explainable Artificial Intelligence: A SurveyarXiv preprint arXiv:2312.12936, 2023
2022
2020
Supervised Concept Learning
Here we include some relevant works in concept representation learning that assume concept-labels are provided in some manner to learn concept representations from which explanations can be then constructed:
2023
2022
- Concept activation regions: A generalized framework for concept-based explanationsAdvances in Neural Information Processing Systems, 2022
- Glancenets: Interpretable, leak-proof concept-based modelsAdvances in Neural Information Processing Systems, 2022
- Concept embedding models: Beyond the accuracy-explainability trade-offAdvances in Neural Information Processing Systems, 2022
2020
2018
- Net2vec: Quantifying and explaining how concepts are encoded by filters in deep neural networksIn Proceedings of the IEEE conference on computer vision and pattern recognition , 2018
- Interpretability beyond feature attribution: Quantitative testing with concept activation vectors (TCAV)In International conference on machine learning , 2018
2017
- Network dissection: Quantifying interpretability of deep visual representationsIn Proceedings of the IEEE conference on computer vision and pattern recognition , 2017
Unsupervised Concept Learning
In contrast to the works above, the following papers attempt to learn concept representations without implicit or explicit concept labels. This is done by the means of concept discovery and represents a particularly active are of reasearch in this field:
2023
-
- Tabcbm: Concept-based interpretable neural networks for tabular dataTransactions on Machine Learning Research, 2023
2021
- Gcexplainer: Human-in-the-loop concept-based explanations for graph neural networks3rd ICML Workshop on Human in the Loop Learning,, 2021
2020
- On completeness-aware concept-based explanations in deep neural networksAdvances in neural information processing systems, 2020
2019
- Towards automatic concept-based explanationsAdvances in neural information processing systems, 2019
2018
- Towards robust interpretability with self-explaining neural networksAdvances in neural information processing systems, 2018
Reasoning with Concepts
Finally, we include some papers that describe very interesting things one can do once one has learnt some concept representations (regardless of whether these representations were learnt with or without concept supervision). These works are highly related to the field of neuro-symbolic reasoning and we discuss them in more detail in our presentation:
2024
- DiConStruct: Causal Concept-based Explanations through Black-Box DistillationarXiv preprint arXiv:2401.08534, 2024
2023
2022
-
- Entropy-based logic explanations of neural networksIn Proceedings of the AAAI Conference on Artificial Intelligence , 2022
- Algorithmic concept-based explainable reasoningIn Proceedings of the AAAI Conference on Artificial Intelligence , 2022
2021
2020
2019
- Explaining classifiers with causal concept effect (cace)arXiv preprint arXiv:1907.07165, 2019
2018
-
- Deepproblog: Neural probabilistic logic programmingAdvances in neural information processing systems, 2018
2016
2014
Concept-Learning Public Codebases
Below we list some concept-based open-sourced libraries. As with our reference material, this is by no means an exaustive list but rather one that contains libraries we have had the chance to interact with in the past. If you wish to include your library here, and it is related to concept-learning, please do not hesitate to contact us and we will include it here.