Learning Series: Machine Learning Engineering and DevOps (every Thursday, 4PM PST)

Date	Apr 30, 2020
Time	06:00 PM EDT
Cost	Free
Online

Description

Overview
This series will focus on Machine Learning Engineering (MLE) and DevOps.
This will be a weekly series, each session about 2 hours.
Detailed agenda here

Intended Audience
Developers, DevOps

When

• Starting on April 09, 2020, 4pm - 6pm PDT
• Repeats every Thursday, weekly

Can't make it to live training?
No worries. Go ahead and sign up and we will send you recordings of sessions.
You can see all the past recordings in the 'Machine Learning Engineering' page and on each session link below.
PreRequisites

• Must have : Development experience
• Nice to have: Python knowledge

What to Bring

• Please bring a reasonably modern laptop (Corporate laptops with overly restrictive firewalls may not work well; Personal laptops are recommended)
• Need to have a Machine Learning Environment setup on your laptop. Please follow this guide
• [nice to have] download our docker image elephantscale/es-training

Session Format

• Each session is about 2 hours
• These will be intensely hands-on

Sessions
(You can access past recordings in the session links below)
Session 1: A Tour of TensorFlow 2 (2020-Apr-09)
We will discuss

• Changes from TF1
• TF Datasets library
• Efficient loading of data
• Using Google Colab to run ML code

more details and recording available here

Session 2 : Speeding up training using GPU and TPU (2020-Apr-16)
How to utilze GPU and TPU to speed up trainng
More details and recording

Session 3: Monitoring training (2020-Apr-23)
We will discuss:

• Using callbacks and TensorBoard to monitor training progress
• Set model to train until to a point (90% accuracy) using callbacks instead of fixed set # of epochs
• Send periodic summary statistics using Slack and/or Twilio
• Notify MLE when training is done!

more details and recording available here

Session 4: Training in the cloud with GPU (2020-Apr-30)

• Choosing GPU resources in the cloud
• Price comparision of different GPU types
• Running a GPU intensive neural net program on the instance; we will use the CIFAR program we developed in the previous session

more details and recording available here

Session 5: Model Serving (2020-May-07)

• Prepping the model for serving
• Inference architecture
• Simple model serving using Python web service
• Model servers: Tensorflow Model Server, AWS Sagemaker
• Serving a model using Tensorflow Model Server
• Inference load balancing and best practices

more details and recording available here


More sessions will be added
Check here for details