(2022) B-Cos Networks: Alignment Is All We Need for Interpretability.
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 transforms in DNNs by our B-cos transform. As we show, a sequence (network) of such transforms induces a single linear transform that faithfully summarises the full model computations. Moreover, the B-cos transform introduces alignment pressure on the weights during optimisation. As a result, those induced linear transforms become highly interpretable and align with task-relevant features. Importantly, the B-cos transform is designed to be compatible with existing architectures and we show that it can easily be integrated into common models such as VGGs, ResNets, InceptionNets, and DenseNets, whilst maintaining similar performance on ImageNet. The resulting explanations are of high visual quality and perform well under quantitative metrics for interpretability. Code available at github.com/moboehle/B-cos.
| Item Type: | Conference or Workshop Item (A Paper) (Paper) |
|---|---|
| Divisions: | Mario Fritz (MF) |
| Conference: | CVPR IEEE Conference on Computer Vision and Pattern Recognition |
| Depositing User: | Tobias Lorenz |
| Date Deposited: | 17 Jun 2022 09:51 |
| Last Modified: | 17 Jun 2022 09:51 |
| Primary Research Area: | NRA1: Trustworthy Information Processing |
| URI: | https://publications.cispa.saarland/id/eprint/3715 |
Actions
Actions (login required)
 |
View Item |