ExplainabilityTransparencyTrustworthinesscs.CV
Learning Bottleneck Concepts in Image Classification
Bowen Wang, Liangzhi Li, Yuta Nakashima, Hajime Nagahara
April 20, 2023
Computer Vision and Pattern Recognition
79 citations
7 influential
Key Takeaways
-
Concept-based explanations are more accessible than pixel-level ones: Rather than highlighting individual pixels that influence a model's decision (which requires expert interpretation), BotCL explains decisions using human-understandable concepts, making AI systems more transparent to non-experts.
-
Models can learn interpretable concepts without explicit labeling: The method uses self-supervision and specialized training techniques to discover meaningful concepts automatically, rather than requiring humans to manually label what concepts should matter—reducing the annotation burden for building interpretable systems.
-
Interpretability and performance can be achieved simultaneously: By redesigning how neural networks process information through a "bottleneck" that forces decisions to rely on learned concepts, the approach demonstrates that explaining model behavior doesn't necessarily require sacrificing accuracy on the classification task.
-
Practical tools for interpretability are becoming available: The authors provide open-source code and a public demo, suggesting that concept-based interpretability methods are moving from academic research toward practical tools that organizations can potentially use to audit and understand their AI systems.
Abstract
Interpreting and explaining the behavior of deep neural networks is critical for many tasks. Explainable AI provides a way to address this challenge, mostly by providing per-pixel relevance to the decision. Yet, interpreting such explanations may require expert knowledge. Some recent attempts toward interpretability adopt a concept-based framework, giving a higher-level relationship between some concepts and model decisions. This paper proposes Bottleneck Concept Learner (BotCL), which represents an image solely by the presence/absence of concepts learned through training over the target task without explicit supervision over the concepts. It uses self-supervision and tailored regularizers so that learned concepts can be human-understandable. Using some image classification tasks as our testbed, we demonstrate BotCL's potential to rebuild neural networks for better interpretability. Code is available at https://github.com/wbw520/BotCL and a simple demo is available at https://botcl.liangzhili.com/.