keynote talks
Yann Lecun and Harry Shum depict a grand road map of AI.
Speaking speed of Lecun is 3x faster than normal people. fail to take and understand most part of his talk. The talk focused on CNN related project.
keyword:
Mask R-CNN instance segmentation
ConvSeq2Seq for traslation
FAIR
memory network
stacked-augmented
neural turing network
differentiable memory
Harry Shum’s talk use XiaoIce as the example to depict the AI’s feature and MSRA’s work.

Computer vision topic
Some application in computer vision. There is no impressive talk.
keyword
pre-train use ImageNet
fine-tuning
gabor filter band
FC layer
Alex net

Security topic
Dawn Song is the distinguished scientist in Secure deep learning. List many aspect that hackers can use in deep learning model and application. Deep learning’s secure should start to be considered.
Other three’s talk isn’t suitable for me. very boring

optimization topic
Mathematics + English + BriefTalk = Though I know the word he talk, can’t understand any of them.

Conclusion
Math related topic need a detailed preview !!!
Most irrelevant application talking has nothing to do with you…
Stay in RUJIA by myself….