技术问题请联系 网站管理员
|
报告信息 |
| 题目：Equivariant Neural Networks (等变深度网络) |
| 报告人：林宙辰 教授，北京大学 |
|
摘要：Equivariance is an important property in many data processing tasks such as image recognition. Deep networks can save a lot of parameters and be more robust if equivariance is taken into account. In this talk, I will present my series work on designing practical equivariant networks, with discussions on symmetry discovery. (等变性是图像识别等诸多数据处理任务中的一项重要性质。深度网络若能融入等变性设计，便可大幅减少参数量，同时提升模型鲁棒性。本次报告中，我将介绍本人在实用等变网络设计方面的系列研究工作，并围绕对称性挖掘展开探讨。) |