Artificial Neural Nets and the Representation of Human Concepts
Not on the map. Why?
Abstract (arXiv)
What do artificial neural networks (ANNs) learn? The machine learning (ML) community shares the narrative that ANNs must develop abstract human concepts to perform complex tasks. Some go even further and believe that these concepts are stored in individual units of the network. Based on current research, I systematically investigate the assumptions underlying this narrative. I conclude that ANNs are indeed capable of performing complex prediction tasks, and that they may learn human and non-human concepts to do so. However, evidence indicates that ANNs do not represent these concepts in individual units.
Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Builds on 33 of its 83 references that are in the map. First posted on arXiv 8 Dec 2023.
Builds on
33 of 83 references.
References · 33
- 2017Attention Is All You Needcited by 59,932
- 2014Adam: A Method for Stochastic Optimizationcited by 36,960
- 2014Explaining and Harnessing Adversarial Examplescited by 7,287
- 2013Intriguing properties of neural networkscited by 5,114
- 2016"Why Should I Trust You?": Explaining the Predictions of Any Classifiercited by 5,040
- 2022Locating and Editing Factual Associations in GPTcited by 2,609
- 2020Shortcut Learning in Deep Neural Networkscited by 1,928
- 2018ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustnesscited by 1,430
- 2017Interpretability Beyond Feature Attribution: Quantitative Testing with Concept Activation Vectors (TCAV)cited by 1,309
- 2020Concept Bottleneck Modelscited by 1,022
- 2019Adversarial Examples Are Not Bugs, They Are Featurescited by 950
- 2018Reconciling modern machine learning practice and the bias-variance trade-offcited by 919
21 more
- 2017Network Dissection: Quantifying Interpretability of Deep Visual Representationscited by 703
- 2018Do Better ImageNet Models Transfer Better?cited by 560
- 2017Adversarial Patchcited by 514
- 2018A Survey on Deep Transfer Learningcited by 417
- 2017The Expressive Power of Neural Networks: A View from the Widthcited by 318
- 2023Sources of Hallucination by Large Language Models on Inference Taskscited by 168
- 2023Can ChatGPT Understand Too? A Comparative Study on ChatGPT and Fine-tuned BERTcited by 155
- 2017Learning to Generate Reviews and Discovering Sentimentcited by 147
- 2022From Attribution Maps to Human-Understandable Explanations through Concept Relevance Propagationcited by 141
- 2021Fit without fear: remarkable mathematical phenomena of deep learning through the prism of interpolationcited by 130
- 2018On the importance of single directions for generalizationcited by 99
- 2016Towards an integration of deep learning and neurosciencecited by 77
- 2021BiaSwap: Removing dataset bias with bias-tailored swapping augmentationcited by 70
- 2020Towards falsifiable interpretability researchcited by 53
- 2020The Intriguing Relation Between Counterfactual Explanations and Adversarial Examplescited by 44
- 2017The power of deeper networks for expressing natural functionscited by 35
- 2019Imperceptible Adversarial Attacks on Tabular Datacited by 35
- 2018Revisiting the Importance of Individual Units in CNNs via Ablationcited by 32
- 2022Scientific Inference With Interpretable Machine Learning: Analyzing Models to Learn About Real-World Phenomenacited by 21
- 2020Selectivity considered harmful: evaluating the causal impact of class selectivity in DNNscited by 15
- 2020On the relationship between class selectivity, dimensionality, and robustnesscited by 1
Built on by
Citing papers per year: 2024: 1; 2025: 1; 2026: 1.
3 papers.