11. Andrew NgAndrew Ng
Rocket engines: Deep Learning driven by scale
1 million
connections
(2007)
CPU
10 million
connections
(2008)
GPU
1 billion
connections
(2011)
Cloud
(many CPUs)
100 billion
connections
(2015)
HPC
(many GPUs)
12. Andrew NgAndrew Ng
A yellow bus driving down a road
with green trees and green grass
in the background.
Living room with white couch and
blue carpeting. The room in the
apartment gets some afternoon sun.
Can a computer understand these pictures?
13. Andrew NgAndrew Ng
Supervised learning (learning from labeled data)
YX
Image Caption
A yellow bus driving down a
road with green trees and
green grass in the background.
15. Andrew NgAndrew Ng
( , )
Learning to answer questions
YX
(Image,Question) Answer
The bus is red.
(公共汽车是红色的)
What is the color
of the bus?
(公共汽车是设么颜色的?)
16. Andrew NgAndrew Ng
What is the
color of the
bus?
Data
(image, question)
The bus is red ....
Learning to Answer Questions
24. Andrew NgAndrew Ng
With 99% accuracy, we could
redesign your cellphone using a
speech interface.
Most people don’t understand the difference
between 95% accuracy and 99% accuracy.
99% is game changing.
25. Andrew NgAndrew Ng
Home appliances
(e.g., TV, microwave,
music player, ….)
Car interfaces Wearables
Speech will transform the Internet of Things
30. Andrew NgAndrew Ng
Why deep learning
Amount of data
Performance Older learning
algorithms
Deep learning
How do data science techniques scale with amount of data?
31. Andrew NgAndrew Ng
The problem of scale: Mobile devices
Image 72 keypoints Face segmentation
Binary size Speed
Desktop model 153MB 1.25 fps
Mobile model 800 KB
(190x reduction)
25 fps
(20x speedup)