Towards Explanation of DNN-based Prediction with Guided Feature Inversion
Authors:
Mengnan Du (Texas A&M University); Ninghao Liu (Texas A&M University); Qingquan Song (Texas A&M University); Xia Hu (Texas A&M University)
Abstract:
While deep neural networks (DNN) have become an effective computational tool, the prediction results are often criticized by the lack of interpretability, which is essential in many real-world applications such as health informatics. Existing attempts based on local interpretations aim to identify relevant features contributing the most to the prediction of DNN by monitoring the neighborhood of a given input. They usually simply ignore the intermediate layers of the DNN that might contain rich information for interpretation. To bridge the gap, in this paper, we propose to investigate a guided feature inversion framework for taking advantage of the deep architectures towards effective interpretation. The proposed framework not only determines the contribution of each feature in the input but also provides insights into the decision-making process of DNN models. By further interacting with the neuron of the target category at the output layer of the DNN, we enforce the interpretation result to be class-discriminative. We apply the proposed interpretation model to different CNN architectures to provide explanations for image data and conduct extensive experiments on ImageNet and PASCAL VOC07 datasets. The interpretation results demonstrate the effectiveness of our proposed framework in providing class-discriminative interpretation for DNN-based prediction.
More on http://www.kdd.org/kdd2018/
Видео Towards Explanation of DNN-based Prediction with Guided Feature Inversion канала KDD2018 video
Mengnan Du (Texas A&M University); Ninghao Liu (Texas A&M University); Qingquan Song (Texas A&M University); Xia Hu (Texas A&M University)
Abstract:
While deep neural networks (DNN) have become an effective computational tool, the prediction results are often criticized by the lack of interpretability, which is essential in many real-world applications such as health informatics. Existing attempts based on local interpretations aim to identify relevant features contributing the most to the prediction of DNN by monitoring the neighborhood of a given input. They usually simply ignore the intermediate layers of the DNN that might contain rich information for interpretation. To bridge the gap, in this paper, we propose to investigate a guided feature inversion framework for taking advantage of the deep architectures towards effective interpretation. The proposed framework not only determines the contribution of each feature in the input but also provides insights into the decision-making process of DNN models. By further interacting with the neuron of the target category at the output layer of the DNN, we enforce the interpretation result to be class-discriminative. We apply the proposed interpretation model to different CNN architectures to provide explanations for image data and conduct extensive experiments on ImageNet and PASCAL VOC07 datasets. The interpretation results demonstrate the effectiveness of our proposed framework in providing class-discriminative interpretation for DNN-based prediction.
More on http://www.kdd.org/kdd2018/
Видео Towards Explanation of DNN-based Prediction with Guided Feature Inversion канала KDD2018 video
Показать
Комментарии отсутствуют
Информация о видео
Другие видео канала
![Online Parameter Selection for Web-based Ranking Problems](https://i.ytimg.com/vi/hBjY_3a30UI/default.jpg)
![Deploying Machine Learning Models for Public Policy: A Framework](https://i.ytimg.com/vi/RQHbC0yPp08/default.jpg)
![Risk Prediction on Electronic Healthcare Records with Prior Medical Knowledge](https://i.ytimg.com/vi/o7vx2MubyGA/default.jpg)
![Fairness of Exposure in Rankings](https://i.ytimg.com/vi/krCkf59VllM/default.jpg)
![Notification Volume Control and Optimization System at Pinterest](https://i.ytimg.com/vi/9VsvQoskhlw/default.jpg)
![Multimodal Sentiment Analysis To Explore the Structure of Emotions](https://i.ytimg.com/vi/hwjDRwlTiD8/default.jpg)
![Tax Fraud Detection for Under-Reporting Declarations Using an Unsupervised Machine Learning Approach](https://i.ytimg.com/vi/ramWrCFfhY4/default.jpg)
![Scalable Spectral Clustering Using Random Binning Features](https://i.ytimg.com/vi/Y27XHuAzibA/default.jpg)
![EvoGraph: An Effective and Efficient Graph Upscaling Method for Preserving Graph Properties](https://i.ytimg.com/vi/XfkpodZDsZM/default.jpg)
![Simultaneous Urban Region Function Discovery and Popularity Estimation Via an Infinite Urbanization](https://i.ytimg.com/vi/ibikb_j6EDE/default.jpg)
![Predicting Estimated Time of Arrival for Commercial Flights](https://i.ytimg.com/vi/EgZnDpLvF60/default.jpg)
![Anatomy of a Privacy-Safe Large-Scale Information Extraction System Over Email](https://i.ytimg.com/vi/XylhLyOw_c4/default.jpg)
![A Data-Driven Three-Layer Algorithm for Split Delivery Vehicle Routing Problem](https://i.ytimg.com/vi/zgMSXR2q52s/default.jpg)
![An Extensible Event Extraction System With Cross-Media Event Resolution](https://i.ytimg.com/vi/Mf7jXaEehc8/default.jpg)
![Efficient Mining of the Most Significant Patterns with Permutation Testing](https://i.ytimg.com/vi/o8udCrpdEVY/default.jpg)
![Towards Mitigating the Class-Imbalance Problem for Partial Label Learning](https://i.ytimg.com/vi/eKahTJB0OY4/default.jpg)
![HeavyGuardian: Separate and Guard Hot Items in Data Streams](https://i.ytimg.com/vi/qen2M95U73U/default.jpg)
![Enhancing Predictive Modeling of Nested Spatial Data through Group-Level Feature Disaggregation](https://i.ytimg.com/vi/7VOxDEpDosc/default.jpg)
![Ranking Distillation: Learning Compact Ranking Models With High Performance for Recommender System](https://i.ytimg.com/vi/lj9tCOrIHjs/default.jpg)
![Gotcha - Sly Malware! Scorpion: A Metagraph2vec Based Malware Detection System](https://i.ytimg.com/vi/sM-vCJHvw9A/default.jpg)