Google APAC
Machine Learning
Expert Day
Linkernetworks - Evan Lin / Benjamin Chen
● Tensorflow Summit Recap
● Google APAC Machine Learning Expert Day
● Our lightening talk (Linker Neworks)
Agenda
Who is Evan Lin
● Daily Work:
○ Linker Networks : Cloud
Architect in
● Community:
○ Co-Organizer in Golang
Taipei User Group
● Habit:
○ Project 52
Tensorflow Summit RECAP
Tensorflow Dev Summit 2017
link
Benjamin Chen
Linker Networks
Data Scientist
Taiwan R User Group
Officer
benjamin0901@gmail.com
After 1.0.0
● 1.0.0
○ XLA
○ pip install tensorflow
○ JAVA API
● 1.1.0
○ Keras 2.0-->tf.contrib.keras
■ tf.keras by TF 1.2
○ tf.estimator
TensorFlow Wide & Deep Learning
Wide Model Deep Model
Memorization Generalization
Revelance Diversity
Deep Model
Generalization
Diversity
Wide Model
Memorization
Relevance
Wide & Deep Model
Classify cucumbers with tensorflow
Classify cucumbers with tensorflow
Japanese Idol with DCGAN (link)
APAC Machine Learning
Expert Day 1
Some Interesting Projects
Deep Learning in your flash drive (link)
Tensorflow example from zero to all (link)
Tensorflow What? (link)
Other lightning talks
● Context and attention extraction / modeling
● NLU and Cognitive Architectures
● [Linker Networks] When Kubernetes meets Tensorflow
● [Linker Networks] Running Distributed Tensorflow with
Jupyter Notebook and Kubernetes
● [Linker Networks] Machine Intelligent Cluster
APAC Machine Learning
Expert Day 2
Tensorflow intro with Codelab (link)
Google Cloud Codelab
Classify Manhattan
Classify MNIST images
Linker Networks
When Kubernetes meets Tensorflow
Machine Intelligence Cluster: Use
Tensorflow to improve Kubernetes● Goal:
○ Kubernetes with Machine
Intelligence
● Role played by ML:
○ Maximize utilization
○ Risk mitigation
● Tools Used:
○ Tensorflow
○ Spark Streaming
Utilization Prediction
- Product: Cluster of Machine
Intelligence, CMI
- Goals:
- Predict CPU and memory trend
- Auto-scaling
- Algorithm: LSTM
- Module: Keras
- Trying to tune threshold
Back to Evan
Eliminate engineering bottlenecks
in Machine Learning
Data Collect Probe & Sensor & Smart GW
Vizualization
Data Process
Data Analysis &
Machine Learning
DC/OS Spark ML Tensorflow
DC/OS
Storage
Cassandra
Kafka (Queueing)
Go/Akka (Connector)
Spark (ETL/Streaming)
D3.js
Scikit Learn R
Interactive
Dashboard
Jupyter Notebook
Zeppelin
ML Job
Scheduler
Chronos
HPC (with
GPU)
server Storage SDNStorage SDN
Analytics Machine Intelligence Platform (AMIP)
: Building deep learning platform on top of
Kubernetes
● Goal:
○ Zero setup for Tensorflow
(private/public cloud)
○ Migrate with Jupyter, TensorBoard
and TensorServing
● Tools Used:
○ Kubernetes
○ Tensorflow
AMIP Architecture
Linker is hiring
Cloud Platform Developer
- Familiar with DCOS/K8S
- Strong DevOps experience
Q&A

Google APAC Machine Learning Expert Day