국방 데이터를 활용한 인셉션 네트워크 파생 이미지 분류 AI의 설명 가능성 연구 = A Study on the Explainability of Inception Network-Derived Image Classification AI Using National Defense Data

다국어 초록 (Multilingual Abstract)

In the last 10 years, AI has made rapid progress, and image classification, in particular, are showing excellentperformance based on deep learning. Nevertheless, due to the nature of deep learning represented by a black box,it is difficult to actually use it in critical decision-making situations such as national defense, autonomous driving,medical care, and finance due to the lack of explainability of judgement results. In order to overcome theselimitations, in this study, a model description algorithm capable of local interpretation was applied to the inceptionnetwork-derived AI to analyze what grounds they made when classifying national defense data. Specifically, weconduct a comparative analysis of explainability based on confidence values by performing LIME analysis from theInception v2_resnet model and verify the similarity between human interpretations and LIME explanations.
Furthermore, by comparing the LIME explanation results through the Top1 output results for Inception v3,Inception v2_resnet, and Xception models, we confirm the feasibility of comparing the efficiency and availabilityof deep learning networks using XAI.

참고문헌 (Reference)

1 Ribeiro, "“Why Should I Trust You?” Explaining the Predictions of any Classifier" 2016

2 CHOLLET, Francois, "Xception: Deep Learning with Depthwise Separable Convolutions" 1251-1258, 2017

3 SIMONYAN, Karen, "Very Deep Convolutional Networks for Large-Scale Image Recognition"

4 SZEGEDY, Christian, "Rethinking the Inception Architecture for Computer Vision" 2818-2826, 2016

5 GOLDSTEIN, Alex, "Peeking Insde the Black Box: Visualizaing Statistical Learning with Plots of Individual Conditional Expectation" 24 (24): 44-65, 2015

6 BINDER, Alexandar, "Layer-Wise Relevance Propagation for Neural Networks with Local Renormalization Layers" Springer International Publishing 63-71, 2016

7 SZEGEDY, Christian, "Inception-v4, Inception ResNet and the Impact of Residual Connections on Learning"

8 KRIZHEVSKY, Alex, "Imagenet Classification with Deep Convolutional Neural Networks" 60 (60): 84-90, 2017

9 SELVARAJU, Ramprasaath R., "Grad-Cam:Visual Explanations from Deep Neural Networks Via Gradient-based Localization" 618-626, 2017

10 SZEGEDY, Christian, "Going Deeper with Convolutions" 1-9, 2015

11 TAN, Mingxing, "Efficientnet: Rethinking Model Scaling for Convolutional Neural Networks" PMLR 6105-6114, 2019

12 HE, Kaiming, "Deep Residual Learning for Image Recognition" 770-778, 2016

13 RUDER, Sebastian, "An Overview of Gradient Descent Optimization Algorithms"

14 LUNDBERG, Scott M., "A Unified Approach to Interpreting Model Predictions" 30 : 2017