6. Dial into meetings
Control In-Room Conferencing Systems
Set room temperature or lighting
Book meeting rooms
Replenish office supplies
Schedule meetings
and much more…
Use your voice to…
7. Assisted Interactions
Intelligent Voice Responses
Chatbots
Funneling of incoming requests …
Hi, how can I help you today?
I want to make a reservation for a
table tonight at 8 p.m.
Great, I can help you with that,
how many of you will come
tonight?
Requests
Intentions
Amazon Lex
Fulfill
automatically
Reroute to correct
funnel
9. A m a z o n S u m e r i a n
The fastest and easiest way to create VR, AR, and 3D experiences
10. Display screen
Consumer
Amazon Lex
AWS Cloud
Mobile Phone /
Tablets
Amazon Sumerian
Amazon Lex listens to the voice, interprets the intent
and responds via an Amazon Sumerian host
12. AR.js is fast. It can
run scenes at 60 fps
even on older
phones
FAST
Get your users
running your app as
soon as they can.
No install!
IMMEDIATE
Open source, JavaScript framework for
building web based Augmented Reality
applications.
Amazon Sumerian and AR.js
Browser Based
Augmented Reality
16. Learn independently with little
supervision
Simulates how our brains learn by
creating artificial "neural networks"
Especially useful in computer vision
– learn from the data with little to no
feature engineering
D e e p L e a r n i n g : D e e p N e u r a l N e t w o r k
17. Convolutional Neural Networks (CNNs)
Connectivity pattern between neurons
resembles the organization of the animal
visual cortex
Inspired by the receptive field of the animal
cortex it exploits local correlations in images
Finds applications in image recognition, facial
recognition, or video analysis
19. SSD – Single Shot MultiBox Detector
downsample
body
class predictor
box predictor
input scale 0 scale 1
class predictor
box predictor
20. Joseph Redmon
YOLO - You Only Look Once
Real-Time Object Detection
Darknet – Open source Neural Networks in C
21. EC2
NVIDIA
Tesla V100 GPUs
P3
Deep Learning AMI
Amazon S3
Darknet
Training Images
Model Output
Infer Model on
Camera
AWS Greengrass
AWS Cloud Edge Location
22. Labelling and annotating the dataset can be an intensive task
Tip: Record a video in multiple settings and lights and extract frames
23. Labelling and annotating the dataset can be an intensive task
Tip: Record a video in multiple settings and lights and extract frames
24. Building ground truth data is an
intensive task
Need human intelligence
to annotate data sets
Facebook AI Research
DigitalGlobe | Radiant
Amazon Mechanical Turk
28. AWS Greengrass ML
Includes pre-built TensorFlow and Apache MXNet package
for all devices powered by Intel Atom, NVIDIA Jetson TX2,
and Raspberry Pi
Makes it easy to deploy your machine learning model from
the cloud to your devices
Infer locally reduces latency and cost to predict - Data is sent
only to the cloud only when it requires additional processing
29. Zhe Cao, Tomas Simon, Shih-
En Wei, Yaser Sheikh, Tomas
Simon, Hanbyul Joo, Iain
Matthews, Varun Ramakrishna,
Takeo Kanade
Carnegie Mellon
Perceptual Computing Lab
31. Fully programmable video camera
Optimized for deep-learning on the device with Apache
MXNet, Caffe, TensorFlow
Tutorials, sample code, examples and pre-built models
Integrated with Amazon Sagemaker for custom models
AWS DeepLens
A deep learning-enabled video camera for developers
33. Monitoring App
Camera (Edge Site) ConsumerAWS Cloud
DatalakeAmazon S3
Amazon Rekognition
Local Camera Compliance footage
(picture & video)
Real-time facial
detection
34. Monitoring App
Camera (Edge Site) ConsumerAWS Cloud
Datalake
Amazon S3
Amazon Rekognition
Local Camera
Local real-time
facial detection
AWS Greengrass
Only forward
relevant footage