Mba-Ai Speech and Natural Language Processing (NLP) Technologies
Mba-Ai Speech and Natural Language Processing (NLP) Technologies
Mba-Ai Speech and Natural Language Processing (NLP) Technologies
3
Schedule: 2019/11/23
Speaker recognition technology
recognition/verification (SR/SV)
Speaker diarization
Natural language processing technologies
Word/sentence/document embedding
NLP applications
Chatbot
Colab demo (if possible)
4
General Readings: Book
5
General Readings: Web / App
McKinsey Insights on AI
An executive’s guide to AI
Notes from the AI frontier: Applications and value
of deep learning
TechEmergence
Similar to McKinsey, more about AI for business
Medium (from Twitter co-founder)
More technical
6
Assessment
Written report: 50%
Business plan
Oral presentation (on exam date): 50%
Group project
7
AI and Business
Top 10 Best Paying Jobs in Gig Economy
[https://www.mckinsey.com/business-functions/mckinsey-analytics/our-insights/an-executives-guide-to-ai]
10
Deep Learning (DL)
DL AI
Deep learning is one of many machine learning
algorithms
Machine learning is one area of AI
But currently the most excitements about AI
are attributed to DL
11
Father of Deep Learning
Alexey Ivakhnenko (1913 – 2007)
Published 1965! 12
Contributors of Deep Learning
DNN:
Geoffrey Hinton (1947 -)
UT / Google Brain
CNN:
Yann LeCun (1960 -)
Bell Labs; NYU; Facebook
LSTM-RNN:
Jürgen Schmidhuber (1963 -)
Dalle Molle Institute for Artificial
Intelligence Research, Switzerland
13
3 Pillars of AI
14
Convergence of Big Data and AI
learning
analysis
http://medicalfuturist.com/
Economist’s Perspective
Current AI is more about prediction
Given an image, predict what objects it contains
Given a speech, predict its words
Prediction is central to decision-making
AI makes prediction cheap
When AI is cheap, AI is everywhere
there will be new businesses
New business strategy
16
Amazon’s Anticipatory Shipping
During online shopping on Amazon, it will offer
you suggestions
Currently, Amazon’s AI predicts what you will
buy at an accuracy of ~5%
What happens if the accuracy goes up to, say,
80%?
Amazon obtained a patent in 2013 on
something called “anticipatory shipping”
17
Credit Card Fraud Detection
What happens when there is fraudulent
transaction on your credit card?
You’d better check your statement every month
Hopefully the transaction will be reversed
Accuracy of fraud detection
1990+: 80%
2000: 90-95%
Today: 98 – 99.9%
• What can card providers do now?
18
AI + HI
Camelyon Grand Challenge 2016
Breast cancer detection from slides of biopsies
Winner: MIT/Harvard’s AI team, 92.5%
Human pathologist: 96.6%
Combined: 99.5% !!
It turns out HI is right when there is cancer; AI
is right when there is no cancer.
19
AI Risks
Liability risks
Algorithms may amplify existing societal biases
Fake news; meddling of 2016 US election
Input data risks: Garbage in, garbage out
Training data risks
Bing copies Google
Feedback data risks
Microsoft’s Twitter chabot named Tay (2016)
20
AI and the World
21