Tencent qianzhiming: Exploration and application of pre training methods in information flow business

Images 、 Text and voice are the most important information transmission carriers for human beings , The research in these fields has also developed rapidly in recent years . Here we mainly focus on image and text perception modeling , The related pre training tasks are briefly described from the perspectives of image pre training and image text multi-modal pre training , And its impact on various content understanding tasks in information flow business , The main contents are as follows ：

• Data as fuel—— Acquisition and empowerment of large-scale pre training data

• Typical construction and evaluation methods of image and multimodal pre training model

• Exploration and practice of image pre training method in information flow service

• Exploration and practice of multimodal pre training method in information flow service