Interpreting Deep Neural Networks through Prototype Factorization

Das, Subhajit; Xu, Panpan; Dai, Zeng; Endert, Alex; Liu, Ren

doi:10.1109/icdmw51313.2020.00068

Cited by 9 publications

(8 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A study involving end users to contrast this approach with other prototype-based methods could potentially lead to new insights regarding the interpretability of the extracted prototypes. Das et al [103] extract prototypes from the latent representation of the input data in a deep neural network. The prototypes are weighted, and a surrogate model is built from the prototypes.…”

Section: Instance-based Explanationsmentioning

confidence: 99%

Explainable AI for Time Series Classification: A Review, Taxonomy and Research Directions

et al. 2022

View full text Add to dashboard Cite

Time series data is increasingly used in a wide range of fields, and it is often relied on in crucial applications and high-stakes decision-making. For instance, sensors generate time series data to recognize different types of anomalies through automatic decision-making systems. Typically, these systems are realized with machine learning models that achieve top-tier performance on time series classification tasks. Unfortunately, the logic behind their prediction is opaque and hard to understand from a human standpoint. Recently, we observed a consistent increase in the development of explanation methods for time series classification justifying the need to structure and review the field. In this work, we (a) present the first extensive literature review on Explainable AI (XAI) for time series classification, (b) categorize the research field through a taxonomy subdividing the methods into time points-based, subsequences-based and instance-based, and (c) identify open research directions research directions regarding the type of explanations and the evaluation of explanations and interpretability.

show abstract

Section: Instance-based Explanationsmentioning

confidence: 99%

Explainable AI for Time Series Classification: A Review, Taxonomy and Research Directions

et al. 2022

View full text Add to dashboard Cite

show abstract

“…Approaches for understanding DNNs typically focus on explaining individual model decisions post-hoc, i.e., they are designed to work on any pre-trained DNN. Examples of this include perturbation-based, [15], [16], [17], activation-based, [18], [19], or backpropagation-based explanations, [5], [20], [21], [22], [23], [24], [25], [26]. In order to obtain explanations for the B-cos networks, we also rely on a backpropagationbased approach.…”

Section: Related Workmentioning

confidence: 99%

B-cos Networks: Alignment is All We Need for Interpretability

Böhle

Fritz

Schiele

2022

2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

View full text Add to dashboard Cite

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.

show abstract

“…Approaches for understanding DNNs typically focus on explaining individual model decisions post-hoc, i.e., they are designed to work on any pre-trained DNN. Examples of this include perturbation-based, [22,27,29], activationbased, [8,17], or backpropagation-based explanations, [3,31,33,34,36,37,39,45]. In order to obtain explanations for the B-cos networks, we also rely on a backpropagationbased approach.…”

Section: Related Workmentioning

confidence: 99%

B-cos Networks: Alignment is All We Need for Interpretability

Böhle¹,

Fritz²,

Schiele³

2022

Preprint

View full text Add to dashboard Cite

Fig. 1. Top: Inputs xi to a B-cos DenseNet-121. Bottom: B-cos network explanation for class c (c: image label). Specifically, we visualise the c-th row of W1→L(xi) as applied by the model, see Eq. ( 13); no masking of the original image is used for these visualisations. For the last 2 images, we also show the explanation for the 2nd most likely class. For details on the visualisation of W1→L(xi), see Sec. 4.

show abstract

Interpreting Deep Neural Networks through Prototype Factorization

Cited by 9 publications

References 17 publications

Explainable AI for Time Series Classification: A Review, Taxonomy and Research Directions

Explainable AI for Time Series Classification: A Review, Taxonomy and Research Directions

B-cos Networks: Alignment is All We Need for Interpretability

B-cos Networks: Alignment is All We Need for Interpretability

Contact Info

Product

Resources

About