AWS News Blog

re:Invent 2020 Liveblog: Machine Learning Keynote

Voiced by Polly

AWS Chief Evangelist Jeff Barr and Developer Advocates Martin Beeby and Steve Roberts liveblogged the first-ever Machine Learning Keynote. Swami Sivasubramanian, VP of Amazon Machine Learning shared the latest developments and launches in Amazon ML/AI, as well as demos of new technology, and insights from customers.

Read the Machine Learning Keynote recap below. You can also read liveblog recaps of Andy Jassy’s re:Invent Keynote, and the re:Invent Partner Keynote.

Stay tuned for our liveblogs of the Infrastructure Keynote on Thursday, Dec. 10 at 7:30 AM (PST) and Werner Vogels Keynote on Tuesday, Dec. 15 at 7:10 AM (PST).

Martin Beeby  10:05 AM
ML to the rescue! It seems like that was what lots of this keynote focused on, giving tools to people like me, people that generally wouldn’t have the skills to do Machine Learning.

Steve Roberts  10:04 AM
Yes, yes I do. But. I only find out about them, and speed and heading, too late to get into position!

Martin Beeby  10:03 AM
I am looking forward to writing the demo when the AWS Panorama Appliance device is GA. Perhaps I could build that Deer detection app that Steve mentioned earlier in the blog? I still can’t believe he has deers and bears in his back yard.

Martin Beeby  10:01 AM
I loved the AWS Panorama Appliance that Matt Wood showed off. We got our hands on the appliance a few weeks ago. I wrote the blog post, but Jeff set it all up and wrote the demo for my blog post: You can read about it here.

Steve Roberts  9:58 AM
The advances in healthcare that ML and AI are driving are personally exciting. My partner is a retired nurse, and when I think back to “how things were” when we started dating, to where we are now, and going, it’s just amazing.

Martin Beeby  9:55 AM
And that’s a wrap. Wow, lots to process there. I liked the look of Amazon SageMaker Clarify in particular. Bias in machine learning models is a common problem that can have huge implications.

Martin Beeby  9:52 AM
We are now watching a cool video about AWS DeepRacer. I tried to build a DeepRacer model at one of the Summits last year. It wasn’t very fast, but it was lots of fun. If you want to try your hand at Machine Learning with DeepRacer then check out this site.

Jeff Barr  9:51 AM
Very cool to see how popular (and fun) DeepRacer has become!

Steve Roberts  9:50 AM
Wow, I did not know F1 had joined in the Deep Racer fun!

Martin Beeby  9:49 AM
Dr. Swami Sivasubramanian, VP of Amazon Machine Learning is now back on stage.

Steve Roberts  9:49 AM
The final tenet: Learn continuously.

Steve Roberts  9:44 AM
“What is quality in the context of healthcare? It’s the confluence of speed, accuracy, and cost.”

Martin Beeby  9:43 AM
Now, on video, we have Elad Benjamin, the General Manager, Radiology Informatics at Philips. Elad is talking about the complexity of dealing with the silos of data in healthcare. He’s presenting a product Philips built called Health Suite, which uses many AWS ML services such as Amazon SageMaker. The product Improves healthcare delivery by analyzing data from those silos and applying ML. Here’s a case study.

Elad Benjamin, the General Manager, Radiology Informatics at Philips is shown on video in a room with lights and a couch.

Martin Beeby  9:38 AM
Announcement: Amazon HealthLake, in preview, uses machine learning models to normalize health data.  Here’s our blog post about it.

Martin Beeby  9:38 AM
Wow, I never realized that Moderna used AWS to help develop the latest Covid-19 vaccine. Doing part of the work in 42 days instead of 20 months.

Steve Roberts  9:33 AM
Now there’s an idea….!

Jeff Barr  9:33 AM
Or shoot water at the squirrels, Steve!

Steve Roberts  9:33 AM
The discussion about the Panorama Appliance, and its SDK, has kicked off all sorts of ideas in my head about using them to monitor for wildlife (deer, bears) in my yard and sending me timely alerts so I can get into position with my camera gear!

Jeff Barr  9:33 AM
I believe that those animated graph structures superimposed on spheres have been seen at re:Invents past.

Jeff Barr  9:31 AM
Matt brings it all together…
Matt Wood stands in front of slide showing Improving industrial processes with machine learning

Steve Roberts  9:31 AM
There’s nothing humble about a #2 pencil. It’s a crucial piece of kit (imo)!

Martin Beeby  9:29 AM
Matt is showing off the new AWS Panorama Appliance. I wrote about this here on the AWS News Blog. It was surprisingly straightforward to use.

Martin Beeby  9:28 AM
Matt is explaining Amazon Lookout for Vision. Marcia wrote about Amazon Lookout for Vision last week on the News Blog and explained how it simplifies defect detection for manufacturing

Jeff Barr  9:25 AM
Monitron looks cool! Perhaps I can use some of those sensors on my A/C, furnace, computers, 3D printer, and so forth.

Martin Beeby 9:25 AM
Matt is speaking about Amazon Monitron. You can read all about it from Julien here on the AWS News Blog. 

Martin Beeby  9:21 AM
Dr. Matt Wood is back on stage talking about how machine learning is reinventing virtually all industries.
Image of Dr. Matt Wood in front of a slide on stage.

Steve Roberts  9:20 AM
Although there’s an obvious use case here for business metrics, I could see Lookout for Metrics also being useful in a developer/devops scenario, to save those daily (hourly?!) inspections of metric graphs “just to see if anything is going wrong.”

Martin Beeby 9:17 AM
Announcement: A preview of a new service called Amazon Lookout for Metrics. It allows you to detect anomalies in your metrics. Here’s our blog post about it.
Swami announces Amazon Lookout for Metrics in front of a slide on stage.

Martin Beeby  9:16 AM
I love the focus on real customer problems. It is easy to get distracted by these technologies’ wow factor and lose sight of the business benefits.

Steve Roberts  9:13 AM
Tenet #4, we need to solve real business problems end-to-end. But, Swami asks, “what makes a good machine learning problem?”

Jeff Barr  9:10 AM
I want QuickSight Q for our blog metrics!

Martin Beeby  9:10 AM
Dorothy Li is now joining us; Dorothy is the VP of Business Intelligence and Analytics here at AWS. Talking about Amazon Quicksight Q. You can read our News Blog on this topic over here.
Dorothy Li speaks in front of a slide about Amazon QuickSight Q on stage.

Steve Roberts  9:09 AM
Amazon Quicksight Q: “Ask business questions in natural language, and get answers instantly.”

Martin Beeby 9:06 AM
Announcement: Amazon Neptune ML bringing ML to graph applications.
Swami announces Amazon Neptune ML in front of a slide on stage.

Martin Beeby  9:05 AM
Announcement: Amazon Redshift ML. Use SQL to make machine learning predictions from your data warehouse.
Swami announces Amazon Redshift ML in front of a slide on stage.

Steve Roberts  9:04 AM
We’re continuing on the theme here of providing the right tools to builders, of all types – data scientists, developers, database experts – to help expand their toolsets and further drive innovation across the org, no matter where the source data or ML use case resides.

Martin Beeby  9:02 AM
Swami is now talking about Amazon Aurora ML; Danilo talked about this last year. You can read about it here on the news blog.

Jeff Barr  9:01 AM
Lots of work to do to add ML to an existing app.

Martin Beeby  8:59 AM
Swami is now talking about how Amazon SageMaker Autopilot is making ML more accessible and expands machine learning to more builders. You can read Julien’s overview about Amazon SageMaker Autopilot from last years event.

Steve Roberts  8:57 AM
Onto tenet #3: good ideas can come from anywhere in the org, so it makes sense to expand ML to more builders

Martin Beeby  8:54 AM
Announcement: Amazon SageMaker Edge Manager, reducing time to get models onto edge devices. Here’s our blog post about it.

Steve Roberts  8:54 AM
Earlier in the keynote Swami spoke about parallelization to process huge models. Now we’re at the opposite end of the scale, at the edge, with devices that don’t have the capacity to handle large models.

Martin Beeby  8:52 AM
Matt Wood using ML to show that Dancing Queen has high danceability. Alexa. Play Dancing Queen!

Steve Roberts  8:52 AM
I suspect if I ran Clarify on a data set representing my music library it would show a considerable bias to less danceability and more ‘prog’, lol.

Martin Beeby  8:50 AM
Jeff, I am not aware of that robot fire story and I feel like I need to be filled in about it later.

Jeff Barr  8:49 AM
Let’s hope that Matt does not set any robots on fire during this demo.

Martin Beeby 8:48 AM
Now it’s Dr. Matt Wood’s turn. Matt is the VP of Artificial Intelligence here at Amazon Web Services.  He is giving as a Demo of the services that were just announced.

Martin Beeby  8:47 AM
Swami is now discussing an announcement from last week. Amazon SageMaker Pipelines, to bring DevOps practices to your machine learning projects.

Martin Beeby  8:45 AM
Announcement: Deep Profiling for SageMaker Debugger enables you to profile machine learning models to identify and fix issues caused by hardware resource usage. Here’s our blog post about it.

Jeff Barr  8:45 AM
No matter how high-level we get, paying attention to low-level resource utilization still pays dividends.

Steve Roberts  8:44 AM
Although it’s obviously not its primary intent, I can see myself using Clarify to help understand how, where. and why predictions are being made from as I continue to dig in and learn more about ML.

Martin Beeby  8:41 AM
Dr. Sephus is talking about how to reduce bias using Amazon SageMaker Clarify. You can read more about it on the news blog too.

Martin Beeby 8:38 AM
Next via video feed is Dr. Nashlie Sephus from the Applied Science team.

Martin Beeby 8:37 AM
Announcement: Amazon SageMaker Clarify helps you detect bias in machine learning models.

Martin Beeby 8:36 AM
Amazon SageMaker Feature Store
allows you to securely store, discover, and share curated data used in training and prediction workflows.

Amazon SageMaker Feature Store, allows you to securely store, discover, and share curated data used in training and prediction workflows.

Martin Beeby  8:34 AM
Swami is now talking about the new SageMaker Data Wrangler to speed up data preparation.

Announcement: Amazon SageMaker Data Wrangler. A new visual interface for Amazon SageMaker lets you prepare data for machine learning applications using a visual interface.

Steve Roberts  8:32 AM
If you make something successful, customers will do more of it – the classic flywheel effect. Swami is highlighting Intuit, one of the early adopters of SageMaker, which has deployed over 50% more models this past year, saving costs and cutting expert review time.

Steve Roberts 8:30 AM
“The future of football together with AWS is very bright.”

Martin Beeby  8:28 AM
The NFL predicting and preventing injury using ML on AWS is a fantastic concept—reviewing data to improve helmet safety and reduce concussions, being just one example that Jennifer has given.

Steve Roberts  8:28 AM
It’s interesting, when we (as fans) talk about insights for sports such as the NFL or F1, we usually think about the on-screen insights to help us understand the game or race. But behind the scenes, there’s a huge amount of additional analysis going on around safety and the discussion we just had about biomechanical analysis around concussion is testament to that.

Steve Roberts  8:25 AM
The insights into games provided by ML has been a huge help to this expat Brit in understanding American Football – go Hawks! – and provides an even bigger benefit to player safety.

Jeff Barr  8:24 AM
Next Gen Stats is cool and impressive (even though I’m not much of a sports fan), but the focus on keeping players safe and healthy does it for me!

Martin Beeby  8:23 AM
Jennifer Langton now joins us; she is the SVP Player Health and Innovation at the NFL.

Jeff Barr  8:23 AM
Some great customer successes — better, faster, cheaper:

Steve Roberts  8:21 AM
Onto tenet #2 – provide the shortest path to success. We aim to do this by providing the tools to help satisfy the need for builders to explore quickly, which is a significant accelerator. In the last year, we’ve released 50 new features for SageMaker for example. This also helps lift the barriers to adoption in what was a complex and costly process.

Martin Beeby  8:20 AM
If you want to know more, here’s our news blog post from Julien Simon.

Jeff Barr  8:19 AM
Being able to train faster is not just for bragging rights. It encourages experimentation and helps builders to get models into production faster than ever before.

Martin Beeby 8:17 AM
Announcement: Managed Data Parallelism in Amazon SageMaker. Train 40% faster. It simplifies training on large datasets that might be as large as hundreds or thousands of gigabytes. Here’s our blog post about it.

Martin Beeby  8:15 AM
Nice stat: 92% of cloud-based TensorFlow runs on AWS, 91% of cloud-based PyTorch runs on AWS.

Jeff Barr 8:14 AM
All of these instance types give customers a lot of choice and put a lot of training and inference power into their hands.

Steve Roberts  8:11 AM
We’re now discussing tenets, starting with firm foundations. Optimized frameworks and infrastructure for training and deployment. This gives builders the freedom to invent.

Martin Beeby  8:11 AM
Builders of all skill levels can unlock the power of machine learning.

Jeff Barr 8:10 AM
Since the beginning, AWS has focused on empowering developers and giving them the freedom to invent.

Jeff Barr 8:07 AM
Over 250 new features per year. This is what we mean when we talk about the “pace of innovation.”

Jeff Barr 8:06 AM
We have a very broad and very deep set of ML/AI offerings, growing in breadth all the time.

Steve Roberts 8:06 AM
Swami is describing how ML is no longer a niche offering. He’s mentioned Dominos using it to meet their goal of 10mins or less for pizza delivery, Roche to accelerate medical experiences, BMW processing 7PB of data with SageMaker, Nike for product recommendations and F1 to analyze over 550M data points on car design and simulations.

Martin Beeby 8:05 AM
Swami is explaining how we have seen incredible customer momentum around Machine Learning. They really have. Things have moved on so much in such a short period. He is explaining how companies like Nike, BMW, and Domino’s are all using ML on AWS.

Dr. Swamin Sivasubramanian on stage

Martin Beeby 8:01 AM
Things are starting here; on stage now, we have Dr. Swami Sivasubramanian, VP of Amazon Machine Learning.

Martin Beeby  7:59 AM
This music from Durante reminds me of that Werner Vogels saying: Dance like nobody’s watching, encrypt like everyone is.

Durante DJing

Martin Beeby  7:51 AM
Hi everyone. I am excited to be liveblogging the first-ever Machine Learning keynote here at AWS re:Invent. We expect Dr. Swami Sivasubramanian to take the stage shortly. I am blogging from my home office in Northampton, England, and looking forward to hearing about all the latest developments in Machine Learning at AWS.

 Durante standing outside at mixing tableMartin Beeby  7:47 AM
It looks like we have Durante providing the music today. That is some setup he has!

AWS News Blog Team

AWS News Blog Team

The AWS News Blog team comprises developer advocates from around the world. It includes Jeff Barr, Danilo Poccia, Marcia Villalba, Channy Yun, Sebastien Stormacq, Antje Barth, Veliswa Boya, Donnie Prakoso, and Irshad Buchh.