2. The AWS ML Stack
Broadest and most complete set of Machine Learning capabilities
VISION SPEECH TEXT SEARCH CHATBOTS PERSONALIZATION FORECASTING FRAUD DEVELOPMENT CONTACT CENTERS
Ground
Truth
Augmented
AI
ML
Marketplace
Neo
Built-in
algorithms
Notebooks Experiments
Model
training &
tuning
Debugger Autopilot
Model
hosting
Model Monitor
Deep Learning
AMIs & Containers
GPUs &
CPUs
Elastic
Inference
Inferentia FPGA
Amazon
Rekognition
Amazon
Polly
Amazon
Transcribe
+Medical
Amazon
Comprehend
+Medical
Amazon
Translate
Amazon
Lex
Amazon
Personalize
Amazon
Forecast
Amazon
Fraud Detector
Amazon
CodeGuru
AI SERVICES
ML SERVICES
ML FRAMEWORKS & INFRASTRUCTURE
Amazon
Textract
Amazon
Kendra
Contact Lens
For Amazon
Connect
SageMaker Studio IDE
NEW
NEW! NEW! NEW! NEW!
NEW!
NEW! NEW! NEW! NEW! NEW!
Amazon SageMaker
3. The AWS ML Stack
Broadest and most complete set of Machine Learning capabilities
VISION SPEECH TEXT SEARCH CHATBOTS PERSONALIZATION FORECASTING FRAUD DEVELOPMENT CONTACT CENTERS
Ground
Truth
Augmented
AI
ML
Marketplace
Neo
Built-in
algorithms
Notebooks Experiments
Model
training &
tuning
Debugger Autopilot
Model
hosting
Model Monitor
Deep Learning
AMIs & Containers
GPUs &
CPUs
Elastic
Inference
Inferentia FPGA
Amazon
Rekognition
Amazon
Polly
Amazon
Transcribe
+Medical
Amazon
Comprehend
+Medical
Amazon
Translate
Amazon
Lex
Amazon
Personalize
Amazon
Forecast
Amazon
Fraud Detector
Amazon
CodeGuru
AI SERVICES
ML SERVICES
ML FRAMEWORKS & INFRASTRUCTURE
Amazon
Textract
Amazon
Kendra
Contact Lens
For Amazon
Connect
SageMaker Studio IDE
NEW
NEW! NEW! NEW! NEW!
NEW!
NEW! NEW! NEW! NEW! NEW!
Amazon SageMaker
14. Business Rules vs Machine Learning
Business Rules look for specific conditions or behaviors
• Business Rules are easily explained and validated
• Sample New Account Registration rule:
ML Models learn more general patterns by looking at lots of examples
• When fraudsters make small tweaks, the model still recognizes them as
suspicious since it’s unlike anything it has seen from legitimate
customers
• ML models are not just good at finding the risky patterns, they’re much
less brittle than rules
If IP_ADDRESS_LOCATION == [’Japan’] and CUST_ADDRESS_COUNTRY == [‘JAPAN’] and CUSTOMER_PHONE_LOC ==
[‘Spain’] THEN Investigate
Prevention Detection
15. Fraud detection is difficult
$$$ billions lost to
fraud each year
Online business prone
to fraud attacks
Bad actors change
tactics often
Rules = more human
reviews
Dependent on others to
update detection logic
16. Fraud detection with ML is also difficult
Top data scientists are
costly & hard to find
One-size-fits-all
models underperform
Often need to
supplement data
Data transformation +
feature engineering
Fraud imbalance =
needle in a haystack
17. Introducing Amazon Fraud Detector
A fraud detection service that makes it easy
for businesses to use machine learning to
detect online fraud in real-time, at scale.
18. Benefits of Amazon Fraud Detector
• Build high quality fraud detection ML models faster
• Stop bad actors at the door
• Built-in online fraud expertise
• Give fraud teams more control
19. Detect common types of online fraud
Designed to help companies detect common types of online fraud
Examples:
• New account fraud
• Online payment fraud (coming soon)
• Guest checkout fraud
• ‘Try Before You Buy’ + post-paid online service abuse
25. Employees spend 20% of
their time looking for
information.
—McKinsey
20%
44%
44% of the time, they cannot find the
information they need to do their job.
—IDC
26. Key Challenges
• Low Accuracy
• 80% of data is unstructured
• Keyword Engines
Complexity
• Scattered Data Silos
• Stale Search Results
• Difficult to set up
27. Impact on Enterprise
• Lower employee productivity
• Increased risk and liability
• Duplication of work
• Creates negative customer experience
30. Amazon Kendra-Rethinking Enterprise Search
Easy to find what
you are looking for
Simple and
quick to set up
Native connectors
Natural language
Queries
NLU and
ML core
Simple API
and console
experiences
Code samples
Continuous
Improvement
Domain
Expertise
38. Use cases
Internal search supporting business functions such as operations,
support, and R&D.
External search helping your customers find the information they need.
CRM, Content Management, and eDiscovery ISVs can build more intelligent
and data-driven applications using Kendra.
40. A day in the life of Lynn
• Lynn is tech lead working on Java projects in an ecommerce company
• part of a distributed development team
• responsible for the backend services (search, order, and shipping) of her company’s high volume
site
• Her responsibilities span the entire application development and operations cycle
• D: We found a data corruption issue in
production.
• L: Let’s find the root cause.
D: I think it is due to a data
race.
Could we have caught it
during code reviews? I wish we had someone
who really understands concurrency.
• O: The site latency is increasing. I just got paged!
• L: Let’s find the root cause.
O: The CPUs are overloaded. Can we increase
the fleet size?
• L: We increased the fleet size last month. The traffic is pretty
much the same. What’s going on?
• O: ???
• L: OK, let’s increase the fleet size.
How do we find out what’s actually going
on? I wish we’ve a performance expert in our team!
41. What’s on Lynn’s mind?
How can we
improve
code quality?
Are we giving
lowest latency
to our
customers?
Are our
infrastructure
costs just
bloating?
43. What’s missing in Lynn’s ecosystem?
• Detection of code defects early in the cycle
• Keeping up with coding best practices
• Identifying performance bottlenecks and linking them to code
• Tools for visualizing application performance
• Availability of expertise
• Faster time to resolution and remediation
• Developers need a truly integrated tool.
• The tool should provide actionable recommendations across
phases in the life cycle.
47. CodeGuru Profiler
• Trained to find high-potential
optimizations
• High Latency
• High CPU Utilization
• Gives Code fix recommendations
48. Amazon Developer Feedback on Profiler
Chris Butterfield, SDE
CodeGuru Profiler’s recommended fix removed the thread contention
which was using 55.97% of CPU time. After the fix a single host could
now serve ~7.5x more traffic than before. We reduced our number of
instances by ~75% while still handling the same traffic
Rajesh Konatham, SDE
After following Profiler’s recommendation to remove these clones, we
saw huge reductions in CPU utilization – a 40% reduction on the
synchronous fleet and 67% reduction on the asynchronous fleet
50. Contact Lens for Amazon Connect
Advanced search Detailed analytics &
sentiment analysis
Automated contact
categorization
Theme detection
(coming soon)
Supervisor assist
(coming soon)
Open and flexible
data
Contact Center Analytics for Amazon Connect powered by Machine Learning
The out-of-the-box experience makes it easy for contact centers and their staff to use the
power of ML with just a few clicks.
57. Input a melody by connecting
the AWS DeepComposer
keyboard
Choose from jazz, rock, pop,
classical, or build your own
custom genre model in Amazon
SageMaker
Publish your tracks to SoundCloud
from the console. Export MIDI files
to your favorite DAW