Pay Attention via Binarization: Enhancing Explainability of Neural Networks via Binarization of Activation

Tashiro, Yuma; Awano, Hiromitsu

doi:10.1109/iscas48785.2022.9937289

Cited by 1 publication

(3 citation statements)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…• Numerical experiments on an automated driving task demonstrates a 1.14× improvements in AUC compared to existing methods. This paper is an extended version of our previously published paper [16]. The enhancements are summarized as follows.…”

Section: Introductionmentioning

confidence: 99%

“…• While [16] required two dedicated convolutional feature extractors, this paper proposes to share a single convolutional feature extractor to reduce the computational complexity.…”

Section: Introductionmentioning

confidence: 99%

“…• While [16] has used floating point representations for network activations, this paper employs discretized activations to improve the explainability.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Pay Attention via Quantization: Enhancing Explainability of Neural Networks via Quantized Activation

Tashiro

Awano

2023

IEEE Access

Self Cite

View full text Add to dashboard Cite

Modern deep learning algorithms comprise highly complex artificial neural networks, making it extremely difficult for humans to track their inference processes. As the social implementation of deep learning progresses, the human and economic losses caused by inference errors are becoming increasingly problematic, making it necessary to develop methods to explain the basis for the decisions of deep learning algorithms. Although an attention mechanism-based method to visualize the regions that contribute to steering angle prediction in an automated driving task has been proposed, its explanatory capability is low. In this paper, we focus on the fact that the importance of each bit in the activation value of a network is biased (i.e., the sign and exponent bits are weighted more heavily than the mantissa bits), which has been overlooked in previous studies. Specifically, this paper quantizes network activations, encouraging important information to be aggregated to the sign bit. Further, we introduce an attention mechanism restricted to the sign bit to improve the explanatory power. Our numerical experiment using the Udacity dataset revealed that the proposed method achieves a 1.14× higher area under curve (AUC) in terms of the deletion metric.

show abstract

Section: Introductionmentioning

confidence: 99%

“…• While [16] required two dedicated convolutional feature extractors, this paper proposes to share a single convolutional feature extractor to reduce the computational complexity.…”

Section: Introductionmentioning

confidence: 99%