B-cos Alignment for Inherently Interpretable CNNs and Vision Transformers

2023/06/19 by Moritz Böhle, Böhle, Moritz, Navdeeppal Singh +5 · 6 citations
Computer Science · #Advanced Neural Network Applications #Adversarial Robustness in Machine Learning #Computer Vision and Pattern Recognition (cs.CV) #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences

paper · pdf · doi:10.48550/arxiv.2306.10898

openalex publication_date 2023/06/19 · openalex created_date 2023/06/22 · openalex updated_date 2026/07/28

Abstract

We present a new direction for increasing the interpretability of deep neural networks (DNNs) by promoting weight-input alignment during training. For this, we propose to replace the linear transformations in DNNs by our novel B-cos transformation. As we show, a sequence (network) of such transformations induces a single linear transformation that faithfully summarises the full model computations. Moreover, the B-cos transformation is designed such that the weights align with relevant signals during optimisation. As a result, those induced linear transformations become highly interpretable and highlight task-relevant features. Importantly, the B-cos transformation is designed to be compatible with existing architectures and we show that it can easily be integrated into virtually all of the latest state of the art models for computer vision - e.g. ResNets, DenseNets, ConvNext models, as well as Vision Transformers - by combining the B-cos-based explanations with normalisation and attention layers, all whilst maintaining similar accuracy on ImageNet. Finally, we show that the resulting explanations are of high visual quality and perform well under quantitative interpretability metrics.

Cited by

• FaCT: Faithful Concept Traces for Explaining Neural Network Decisions
• Explaining Similarity in Vision-Language Encoders with Weighted Banzhaf Interactions
• Pulling Back the Curtain on Deep Networks
• Attri-Net: A Globally and Locally Inherently Interpretable Model for Multi-Label Classification Using Class-Specific Counterfactuals
• Interpretable Diffusion Models with B-cos Networks
• PiPViT: Patch-based Visual Interpretable Prototypes for Retinal Image Analysis

Related

• B-cos Networks: Alignment is All We Need for Interpretability
• B-cosification: Transforming Deep Neural Networks to be Inherently Interpretable
• B-cos LM: Efficiently Transforming Pre-trained Language Models for Improved Explainability
• Towards interpreting computer vision based on transformation invariant optimization
• Convolutional Dynamic Alignment Networks for Interpretable\n Classifications