2. Why Amazon?
• #1 Cloud Platform
• Company as a platform – gradually expanding integrated tool set
• Have business and consumer applications on top, not just pure developer tools
• Strong engineering culture
• Consistent product releases
4. AWS Language services
Sophisticated services oriented at language processing
Translation, text to speech, speech to text and analysis
Integrated with other services
Reactive and streaming
6. AWS Lex
• Conversational interface for text and voice
• Same technology as Alexa
• Cross messenger, platform, channel
• Easily extendible via Lambda
• Fits into call center scenario
• Enterprise integration
7. Apache MXNet
Deep Learning library backed by Amazon and Microsoft. Direct
competitor to Google TensorFlow
• Better multi GPU performance
• More clean API
• Real life model deployment is easier
• More efficient and smaller resulting model footprint
8. Gluon
High level API on top of Apache MXNet. Direct competitor to Keras
• Imperative style execution (Symbolic – Imperative – Hybrid)
• Symbolic API for simpler model definition
• Flexible data loading and transformation
10. AWS SageMaker
Why?
• Doing real ML is time
consuming
• Team effort
• Prototype!=Production
When?
• For real projects
• Medium and big
computational loads
• With process in place
What?
• Organizes ML development
• Integrated environment
with cloud infra
• Rapid and complex
development
11. AWS SageMaker
Build
• Jupyther
• Spark
• Point & Click
• Custom
Train
• Zero set up time
• Streaming & distributed
• Docker & ECS
• Local or other platforms
Host
• One click deployment
• Cloud perks
• A/B testing
• Use your own
12. AWS Greengrass
Why?
• Latency is important
• Bandwidth a problem – too
much data or too slow
• Availability of device, cloud is
not everywhere
• Privacy of data access
When?
• Connected devices
• Robotics
• Healthcare
• Smart City
• Industrial Machinery
What?
• Extends the cloud to the
device
• Basically an Edge compute
scenario
• Local managed container with
all key parts in place
13. AWS Greengrass
Car
Train on the cloud
More data
More computing
Deploy to the edge
Local Lambda
Local broker, security and shadow
Minimum:
Single core CPU 1 Ghz
128MB Ram
x86 and ARM
Linux (Ubuntu or Amazon)
Pre-build:
Intel Atom
NVIDIA Jetson TX2
Raspbery Pi
Windfarm House Factory
17. Key takeaways
• Amazon Machine Learning consist of well integrated blocks for end to end
development of any complexity and budget
• Amazon was late to the game with developer frameworks but quickly
catching up
• Innovating not only on frameworks level: Edge, Conversational, Tools,
Integrations
• Already interesting alternative to Google ML tools
https://aws.amazon.com/mxnet/
https://www.lohika.com/blog/
https://aws.amazon.com/mxnet/
https://www.linkedin.com/in/zenykmatchyshyn/