Artificial Intelligence: Case studies (what can you build)

In this presentation, I show what and how you can build AI products in short, mid and long term.

1. Artificial Intelligence:
Case studies
Rudradeb Mitra
https://www.linkedin.com/in/mitrar/

2. My objective
Make you guys build something in AI

3. Two kinds of algorithms
•Deep learning (supervised or unsupervised) with Neural
networks
•Reinforcement learning

4. Deep learning with NN

5. What can you build?
•Short term - Things you can build from a month to 6
months using existing tools. Mostly feature.
•Mid term - Things you can build within 1-2 years. Can use
some existing frameworks but you also need to add an
app.
•Long term - Solving some visionary major problems.

6. Short term
•Chat bots - Language analysis, sentiment analysis..
✓pick up a sector, a customer and build their bot
Trained Neural Network
Interface /API
You

7. How to build a simple chatbot

8. How to build a simple chatbot
User Says "From where can I buy burgers near Times Square in
Manhattan".

9. Short term
Use existing tools and available softwares.
•Chat bots - Language analysis, sentiment analysis..
✓pick up a sector, a customer and build their bot
•Image recognition
✓used by e-commerce (search, advance AR)

10. Reinventing shopping
experience
Google cloud vision API

11. Mid term
Your Data
Interface /API
Neural Network
Existing Data

14. Mid term
•Reduce road accidents by driver analysis
✓Insurance companies
•Predict customer buying behavior
✓Retail
•Predictive maintenance
✓Industrial IoT

15. Identifying risky drivers

17. Classification with NN
Driving score
Mobile
usage
score

18. Predicting customer buying
behaviors

19. vdts
product_iduser
_id
eval_s
et
order_num
ber
order_do
w
order_hour_of_
day
days_since_prior_or
der
2539329 1prior 1 2 8
2398795 1prior 2 5 7 15
473747 1prior 3 7 12 20
22544786 1prior 4 1 7 21
4215438 1prior 5 3 15 28
2295261 1prior 6 2 7 19
2295261 1prior 7 6 20 20
2550362 1prior 8 5 14 14
1187899 1prior 9 2 16 0
2168274 1prior 10 2 8 30
1501582 1train 11 1 11 10
Instacart Data - 2M order data

20. 2539329
2398785
473747
2550368
1187899
(Product ID)
Predicted product ID
• Our orders follow some pattern
Existing approach

21. Order time

22. Time 2Time 1 Time 3
LSTM (Recurrent Neural
Netowrk)

23. Time 1 Time 2
LSTM - Memory
TensorFlow library for LSTM

24. Predictive maintenance -
Industrial IoT

25. Long term Visionary
problems
•Approach to AI - Intuition building
•Future algorithms?

26. Reinforcement learning:
Building intuition

28. •AlphaGo was able to build
intuition.
•Train with NN and then let it
play with itself.
•Learning through
Reinforcement learning.
Intuition in machines

29. Future AI algorithms
•How can we assure that AI/AS are accountable?
•How can we ensure that AI/AS are transparent?
•How can we extend the benefits and minimize the risks of
AI/AS technology being misused?

30. – Eliezer Yudkowsky
“By far the greatest danger of Artificial Intelligence
is that people conclude too early that they
understand it.”