Developer Tech Minutes: Swin Transformer

Описание: Microsoft researchers developed Swin Transformer, a new model architecture that has significantly outperformed previous dominant convolutional neural networks on broad vision problems. In version 2 of Swin Transformer, they then joined this new architecture with a new learning approach, known as SimMIM, or “A Simple Framework for Masked Image Modeling.”

In this Tech Minutes episode, discover how they are working towards unified modeling and learning in AI with Swin Transformers and SimMIM respectively.

Presented by Han Hu, Principal Researcher and Principal Research Manager from Microsoft Research Asia.

0:00 - Introduction of Han Hu and unified architectures and learning approaches
2:01 - Deep dive into Swin Transformer and the convergence of computer vision and NLP
7:25 - Masked Image Modeling and Unified Learning
9:50 - Closing remarks and resources

To learn more, check out the GitHub (https://github.com/microsoft/Swin-Tra...) or the latest research publication (https://aka.ms/SwinV2-paper).

You can also visit our Tech Minutes webpage for additional resources: https://innovation.microsoft.com/en-u...

Finally, you can discover all the latest Tech Minutes for Developers through our YouTube playlist (   • Innovation Tech Minutes  ) or the Innovation Tech Hub (https://innovation.microsoft.com/en-u...)