ResNeXt

ResNeXt is discussed in 133. 对谢赛宁的7小时马拉松访谈：世界模型、逃出硅谷、AMI Labs、两次拒绝Ilya、杨立昆、李飞飞和42 as a FAIR-period computer-vision architecture built by Xie Saining and Kaiming He around the ImageNet Challenge context. Xie treats it as a scalable Representation Learning framework and connects its grouped/cardinality idea loosely to later mixture-style thinking.

Connections

• Xie Saining and Kaiming He — researchers connected to the work in the source.
• FAIR, Meta, and ImageNet — lab, company, and benchmark context.
• Representation Learning and Frontier Model Scaling — architecture and scaling frame.
• Research Taste — source method behind the work.