Febrary 8th Thursday

The THIRTEENTH SV_EXPERIENCE (Febrary 8th Thursday)

• Title : Solving Alpha Go Zero + TensorFlow, Kubernetes-based Serverless AI Models on GPU
• Date and Time : Thursday, 8 Feb 2018, 18:00 am ~ 20:15 pm
• Place : 1355 Market St – Floor 6 San Francisco, CA 94103
• Attendants : GaeulGo, Minjun Koo

Contents

1. Agenda

1. Drinks & light
2. Talk 0 : Meetup Updates and Announcements by Chris Fregly, Founder & Engineer @ PipelineAI
3. Talk 1 : Deploying Serverless TensorFlow AI Models and Functions on a Kubernetes Cluster using PipelineAI and OpenFaaS by Chris Fregly, Founder & Engineer @ PipelineAI
4. Talk 2 : ensorFlow’s New(ish) Estimator and Dataset APIs with PipelineAI Community Edition (http://community.pipeline.ai) by Chris Fregly, Founder & Engineer @ PipelineAI
5. Talk 3 : Alpha Go Zero/AlphaZero with TensorFlow, Probabilistic Methods, and Neural Network Techniques by Brett Koonce (https://www.linkedin.com/in/brettkoonce), CTO @ Quarkworks (https://quarkworks.net/)

2. What did we listen in this meet-up

In Talk 3,

• ( I referenced ‘http://sanghyukchun.github.io/97/’ when I heard the presentation. It explains in Korean.)
• The speaker used play chess when he attended school. He watched alphago and felt interested.
• He talked about the overview of alpha go. The order is ‘discussing game and rules’,’uct+random rollouts -> MCTS’, ‘MCTS + policy + value -> Alpha Go’ and finally Alpha go zero came out.
• First of all, he explained the rule of Go ( called ‘baduk’ in Asia).
• Secondly, he explained the algorithm called UCT(Upper Confidence Bounds for Trees).
• Thirdly, he talked about ‘random rollouts’.
• Fourth, he talked about ‘alpha go’ and the policy network.
• Finally, he talked about ‘alpha go zero’. It use single network to predict best move and winning odds. ( refer : https://applied-data.science/static/main/res/alpha_go_zero_cheat_sheet.png )
• He also talked ‘alpha zero’ and it teaches self and study.
• (We can play alpha go zero in here : http://www.alphago-games.com/ )

( We ate dinner here. Koo and I thought if we stay in SF, we would attend meetup every night to meet people and eat dinner. )

• What was different from before, we talked with 2 people when we eat dinner. One was majoring data science and the other was interested in playing alpha go.
• We talked about the environment between Korea and US. We have many thing to say because we experienced directly.
• We also felt ‘we’ve learned so much because we can talk other people what we’re doing’.

4. Talk our feelings.

• Gaeul Go : It was quite difficult to me because I don’t know well about AI actually. However, it was also important experience because I can experience
• Minjun Koo : I was surprised to see the AI I learned at school.