arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2106.09982cs.CV
收藏

基于变换不变优化的计算机视觉解释

Towards interpreting computer vision based on transformation invariant optimization

State Key Laboratory of High-end Server & Storage Technology(高端服务器与存储技术国家重点实验室);Shandong Hailiang Information Technology Institutes(山东海亮信息技术研究院)

Chen Li, Jinzhe Jiang, Xin Zhang, Tonghuan Zhang, Yaqian Zhao, Dongdong Jiang, RenGang Li

摘要:

解释深度神经网络(DNN)如何进行预测是人工智能的一个重要领域,这阻碍了DNN的广泛应用。学习表征的可视化有助于我们人类理解DNN的愿景。在这项工作中,可视化的图像,可以激活神经网络的目标类生成的反向传播方法。在图像生成过程中,通过旋转和缩放操作引入变换不变性,使图像的可视化效果得到显著改善。最后,我们展示了一些案例,这种方法可以帮助我们深入了解神经网络。

英文摘要:

Interpreting how does deep neural networks (DNNs) make predictions is a vital field in artificial intelligence, which hinders wide applications of DNNs. Visualization of learned representations helps we humans understand the vision of DNNs. In this work, visualized images that can activate the neural network to the target classes are generated by back-propagation method. Here, rotation and scaling operations are applied to introduce the transformation invariance in the image generating process, which we find a significant improvement on visualization effect. Finally, we show some cases that such method can help us to gain insight into neural networks.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑