2015 DL review
Rebooting AI

RL area

Beginner friendly: OpenAir

NLP

2017, 2018,2019 is the transformer winning years -Megatron LM from nvidia -XLnet: Mutation of bert: outperform the vanilla bert -ALBERT light way bert

ACL winner: Multiple domains dialogue

Alexa prize about open-domain conversation

break conversations into pieces
connected small pieces, be prepared for the back& forward connection.
Pay right attentions on important part
Create opinion
ML
Make fun but correction. Informative data isn’t the primary goal here.

Code2Seq

RL

Multiagent-learning

Imperfect information game

Science of Deep Learning

lottery hypothesis of DL: very small piece of network can replace whole networks.
graphic deep learning.

Active learning pipeline(aka data engineering)

create dump neural network and continue to improve all the time.