This year re:Invent has been quite far from usual: firstly, the conference is entirely online. Secondly, it is lasting three weeks in a row, with a bonus update scheduled for mid-January. Bonus point, we just had a dedicated Keynote entirely focused on machine learning delivered by Swami Sivasubramanian, VP Amazon AI.
Since talking about machine learning is super cool, and AWS is announcing a lot of new features and services, Swami is not the only keynote with AI stuff in it: Andy Jassy delivered a couple of mind-blowing news last Tuesday, and Dr. Matt Wood joined Swami to further deep dive into presented topics.
Being an AWS Hero, with interests in machine learning and serverless, makes my job wonderful these days.
Here a few reactions of fellow heroes:
A lot of features have been announced related to SageMaker. As Mike Chambers pointed out, it appears clear that AWS is trying to onboard data scientists with no prior cloud expertise and provide machine learning capabilities to developers without any data science knowledge.
One step towards this ambitious goal has been the announcement last week of SageMaker Data Wrangler, SageMaker Feature Store, and **SageMaker Pipelines.
This week, Swami presented another interesting feature, focused on model bias detection and explainability: Amazon SageMaker Clarify, which has been received with enthusiasm:
"It helps you to detect biases in your dataset. I think it's fantastic; it's just surfacing that this problem exists. A lot of people don't even realize this a problem at all. It's pretty amazing!"
This feature could improve a lot the application of machine learning to many fields:
"I think it's a big step forward getting people thinking about ethics and AI and responsible AI"
Another common issue when dealing with machine learning models is related to profiling and debugging. Here comes Amazon Deep Profiling for SageMaker Debugger, which offers the capability to drill down into how a model performs with his focus centered on visibility.
" Visibility is key. Machine learning, especially when you are starting, is a kind of magic black box. I'd like things providing this sense of how things work"
Another enabling technology is easing things out for data scientists when dealing with deployments (or the so-called MLOps). Sagemaker Pipelines
_"AWS is surfacing services that are elsewhere and make them available within SageMaker."
In this direction moves also Amazon SageMaker Edge Manager that aims to simplify deployments to edge devices from the cloud.
"Hopefully now you can actually manage huge fleet of edge (deep learning) devices"
In an effort to remove friction to developers using ML where data is, today AWS announced the support of its machine learning bridge for two additional databases: Amazon Redshift ML and Amazon Neptune ML.
A re:Invent keynote couldn't be over without the announcement of completely new services, and there they came: Amazon Lookout for Metrics complements AWS solution stack in the anomaly detection domain, adding a flexible service for time series analysis which is probably at the core of Amazon Lookout for Equipment, Amazon Lookout for Vision, and Amazon Monitron. Lookout for metrics exposes the capability to track multi-dimensional metrics and how they change through time.
"We've building anomaly detection systems for decades, but this makes things easier because you essentially need just to have an anomaly and figure out to detect it, but it was a lot of back and forth. But now it's just magic."
" Applying ML to real-world use cases such as gluing a Lookout sensor to a manufacturing machine or connecting a Panorama appliance to a network of existing cameras is the right way to get ML to be useful for more people and companies"
AWS Heroes are amazed by the direction AWS is leading, pushing data scientists to the cloud and developers without prior expertise to the complex domain of artificial intelligence.
"What we have seen so far in reInvent 2020, and what we see again in Swami's keynote, is AWS focusing on getting real business and social value out of ML and AI, not just jumping on the latest shiny thing. We see this with practical end-to-end hardware powered solutions like Monitron and Panorama, through to high level enabling announcements like Amazon HealthLake."
This year machine learning keynote saw the rise of Amazon SageMaker Studio as a complete IDE for data scientists and the deployment of vertical solutions addressing industrial domains such as healthcare and industrial anomaly detection. It means AI and machine learning are becoming mainstream components in any real-world application, not just a promising research field. In this scenario, serverless or managed services offer a straightforward and clear advantage compared to on-premise deployment, either for model training, operation, and no model at all, which represent the real competitor to players such as AWS.