AWS Machine Learning keynote live blog re:Invent 2020

Dec 8, 2020 | 4:45 PM — 7:00 PM CET

This is the first time AWS has introduced a keynote for machine learning(ML).

4:45–5 PM → Some great music by White Buffalo before the keynote

5 PM → VP of machine learning DR Swami Sivasubramanian started an introduction about ML, he explained the three layers that ML offers:

1- ML frameworks & infrastructure

2- Amazon SageMaker

3- AI Services (6 more launch only this year)

Provide a firm foundation gives you the freedom to reinvent said Dr. Swami Sivasubramanian

Three primary frameworks for ML

90% of Machine Learning runs on TensorFlow and PyTorch. Its important to grow in this area, helping builder to learn faster said Dr. Swami Sivasubramanian

Below the list of new machine learning services and features!

AWS Inferentia (lower cost inference in the cloud)

• 45% lower costs per interface than comparable GPU based instances
• 30% higher throughput than comparable GPU based instances
• Amazon Alexa archives 25% lower end to end latency

AWS Tranium (New machine learning chip customer. designed by AWS to deliver the most cost-effective training in the cloud)

• Most teraflops of any ML instances in the cloud
• Use same Neuron SDK as Inferentia
• Available as Ec2 Instances or Image SageMaker

[New Launch] Faster Distributed Training on Amazon SageMaker 40% faster training.

Machine learning is one of the services grows faster in 2020 in AWS said: Dr. Swami Sivasubramanian

[New Launch] Amazon SageMaker Data wrangeler

The faster way to prepare data for Ml, training data without a single line of code.

[New Launch] Amazon SageMaker Clarify

• Train data for tv show recommendation
• Data labeling and selection
• Model drift

CPU is one of the most expensive said: Dr. Swami Sivasubramanian

[New Launch] Deep Profiling for SageMaker Debugger

Debug resources help to identify bottlenecks, to detect bias in machine learning workflow.

• Visualize different systems resources including GPU, CPU, and I/O memory.
• Analyze resources utilization and get recommendations on how to adjust
• Offline training jobs to run at any point in the machine learning workflow

[New Launch] Amazon SageMaker Pipeline

• Seamlessly manage each step of your end to end workflow
• Share end re-run workflow
• Get started quickly with a preconfigured customizable workflow template
• Compare workflows visually to optimize model performance

We need a lot of different types of data, then Sagemaker can use this data to training the model said Matt Wood AWS VP Artificial Intelligence

[New Launch] Amazon SageMaker Edge Manager

Manage and Monitor ML Model across fleets of Smart Devices

[New Launch] Amazon Aurora ML (Add ML based predeicitons to applicaiton via SQL)

[New Launch] Amazon Athena ML(add ML based predeicitons toyour queries on S3) → Invovle ML model form SQL queries)

[New Launch] Amazon Redshift ML(Use SQL to make ML prediciotns form your data warehaouse)

[New Launch] Amazon Neptune ML -> predciting for graph application

[New Launch] Amazon Quick Sight Q(Powerful out of the box ML capabilties)

• Uncover hidden ML insights
• Predict growth and business trends
• Use auto narratives to tell the story of your dashboard

Challenging of Anomaly detection in metrics

[New Launch] Amazon Lookout for Metrics

• Hight latency of detection
• Failed detection
• Too many false detections
• Lack of actionable restuls
• Adaptability to changes in data streams

[Andy Jessy New Launch] Amazon Monitron

[Andy Jessy New Launch] Amazon Lookout for Equipment

[Andy Jessy New Launch] Amazon Panorama Appliance

• Ip62 rated ruggedized device, dust and water-resistant
• 1U and half-rack wide with chassis points
• Multiple Gigabit Ethernet ports dor redundancy or connection cameras from multiple subnets

[Andy Jessy New Launch] Amazon Lookout for Vision

[New Launch] Amazon HealthLake (store, transform and analzye health and lie scencies data in the cloud, or petabyte scale)

• Seamlessly transforms data to understand and extract meaningful medical information
• Organized data in chronological order so that you can look at trends
• Build-in data query, search, and ML capabilities
• Support interoperability, standards like FHIR( Fast interoperability resources) to enable data sharing health systems