北京雁栖湖应用数学研究院

焦小沛

宋丛威

2023年10月26日 15:00 至 16:30

A3-4-312

Zoom 787 662 9899 (BIMSA)

Transformer is a powerful architecture that achieves superior performance on various sequence learning tasks, including neural machine translation, language understanding, and so on. As the core of the architecture, the self-attention mechanism is a kind of kernel smoothing method, or "local model" by the speaker's word. The whole architecure also could be seen as a sequence model of meanshift algorithm that is a classic clustering method. The report aims to give a brief introduction to Transformer for the researchers who benefit from it as soon as possible.

宋丛威于2011年在浙江大学理学院取得应用数学硕士学位，于2014年在浙江大学数学系取得基础数学博士学位，2014-2021年在浙江工业大学之江学院任讲师，2021至今任BIMSA助理研究员。主要研究方向：小波分析，调和分析，机器学习。