HyDRA: hypergradient data relevance analysis for interpreting deep neural networks.

• المؤلفون الرئيسيون: Chen, Yuanyuan, Li, Boyang, Yu, Han, Wu, Pengcheng, Miao, Chunyan
• مؤلفون آخرون: School of Computer Science and Engineering
• التنسيق: Conference or Workshop Item
• اللغة: English
• منشور في: 2021
• الموضوعات: Interpretability; Neural Network; Machine Learning; Artificial Intelligence
• الوصول للمادة أونلاين: https://aaai.org/Conferences/AAAI-21/; https://hdl.handle.net/10356/147652

The behaviors of deep neural networks (DNNs) are notoriously resistant to human interpretations. In this paper, we propose Hypergradient Data Relevance Analysis, or HYDRA, which interprets the predictions made by DNNs as effects of their training data. Existing approaches generally estimate data con...