This was a busy week for all of us at Facebook, as we delivered a broad spectrum of news and tools to the developer community during F8. One of the most exciting news for the machine learning and AI community was the open source announcement of Caffe2, a lightweight and scalable deep learning framework.
Similar to classical computer science fields such as programming language and computer systems, AI today is seeing significant amounts of research and engineering investments. This is usually a sign of extremely rapid progress of the field. We have already seen great interest from the research community in the framework, and are looking forward to receiving contributions and results from researchers using it.
Post F8, we extended an invitation to several university professors to join us for a small workshop to learn more about how Facebook uses Caffe2 in production and the research behind it to solve challenging problems in AI.
The workshop delivered a lively discussion amongst the professors and Facebook researchers & engineers working across various areas such as: Computer Vision, Speech, NLP, HPC, and mobile.
“Caffe2 appears to be a promising deep learning framework that has already proven itself in diverse applications within Facebook, and has potential for broad use externally,” said Srinivas Aluru, Professor and co-Executive Director of the Institute for Data Engineering and Science (IDEaS) at Georgia Tech.
“Having benefited from software rooted from academia like Caffe and Torch, we are extremely thrilled to have the opportunity to contribute to the open source community, and work with academia to push the the state of the art in AI research and products,” said Yangqing Jia, research scientist on the Caffe2 team at Facebook.
Many of the suggestions we heard during the workshop will help us improve Caffe2 further, and hopefully result in long-term collaborations with academia for the benefit of the entire AI community.