Application of Artificial Intelligence in the real World
1
twentybn Quick Introduction
2
● TECHNOLOGY PROVIDER, enabler of A.I. applications, and DEEP LEARNING innovator ● We love progress-by-engineering and believe in RESEARCH-LEAD DEVELOPMENT ● Founded in 03/2015 by 4 MACHINE LEARNING EXPERTS with 15 years of experiences each ● By end of the year 12 incredible A.I. engineers, researchers, and code ninjas ● $ 2.5M seed round from a US-based investor in 03/2016 ● BERLIN OFFICE: Headquarter, Product ● TORONTO OFFICE: R&D
In A Nutshell
● Advisors: ○ PROF. YOSHUA BENGIO (Univ. Montreal)
3
Applications
A.I. that works
4
A.I. that works
Applications
On audio data
5
Application:
Applications
#1 Speech Recognition
6
Application:
Applications
#1 Speech Recognition
7
Application:
Applications
#1 Speech Recognition
8
A.I. that works
Applications
On image data
9
Application:
Applications
#2 Object Detection
10
Application:
Applications
#3 Visual Search
slyce.it 11
Application:
Applications
#4 (Semi-)Autonomous Driving
MobileEye 12
A.I. that works
Applications
On text data
13
Application:
Applications
#5 Machine Translation
AirBnB
14
Application:
Applications
#5 Machine Translation (getting better and better ;)
AirBnB / Google 15
Application:
Applications
#6 Image to Text
Carnegie Mellon University16
A.I. that works
Applications
Beyond “simple” processing
17
Application:
Applications
#7 Image Generation
University of Freiburg 18
Application:
Applications
#8 Game playing
Google DeepMind 19
A.I. that does not work
Applications
(yet)
20
Application:
Applications
#9 Chatbots
21
Application:
Hype Alert!
Applications
#9 Chatbots
22
Application:
Applications
#10 Visually Guided Robotics
FANUC
23
Application:
Applications
#11 Surveillance
24
So what’s the link between all this?
25
Machine Learning
Machine Learning!
26
Machine Learning! Machine Learning
(More specifically: Deep Learning)
27
Principles of Machine Learning 2nd Step: Online Prediction with a (Neural Network) Model
Machine Learning
Input
Trained Model
Output
Elephant
28
Principles of Machine Learning 1st Step: Offline (Supervised) Training of a Model Learning Algorithm
Machine Learning
Labeled Dataset
Trained Model
29
A bit of history of Deep Learning Connecting old and new technology is the key
Machine Learning
Biology
Big Data
1958 Perceptron
1986 Backprop-Algorithm
2011 “Deep Learning”
Mathematic
Hardware (GPUs)
30
A new software development process (Employing Machine Learning)
31
Traditional Software Development Process
ML Software Development Process
Domain experts are highly involved in programming.
Designing
Programming
Integrating
32
Machine Learning Software Development Process
New Software Development Process
Domain and ML experts define data requirements.
Designing
Data Acquisition
Training
Integrating
33
Summary Take home messages
34
Complex Audio, Image and Text Analysis is possible. Machine Learning makes it intelligent. Deep Learning is the driving force. Data acquisition is a requirement and ML expertise is key.
In short:
A.I. works!
Summary
Domain expert programmers are replaced by ML-knowable data curators.
35
Thank you! DR. FLORIAN HOPPE MANAGING DIRECTOR
Contact
[email protected] @Florian_Hoppe
36
THE FOUNDERS
Additional Slides
37
THE FOUNDERS THE FOUNDERS
The humans behind the A.I.
38
DR. CHRISTIAN THURAU CHIEF TECHNICAL OFFICER
DR. FLORIAN HOPPE MANAGING DIRECTOR
DR. INGO BAX
MANAGING DIRECTOR
THE FOUNDERS
DR. ROLAND MEMISEVIC CHIEF SCIENTIST
*from left to right 39
SAAS/DEMO PLATFORM
OUR SAAS/DEMO PLATFORM: powered by the twentybn CORTEX
40
SAAS/DEMO PLATFORM
●
Access our pretrained Neural Networks via REST-API
●
Train your own classifiers
●
Build your own applications on top of our API
●
Wrappers for common languages in development
●
The power of transfer learning: tailor-made solutions for your company using the twentybn CORTEX
●
Ready to be used on-premise and (soon) embedded system 41