Presentation Highlights:
- Why AI and why now?
- Filter through the buzzwords and hype, and
- How to navigate the AI space as a professional?
Authors and Speakers:
- Anurag Bhatia, Sr. Machine Learning Engineer, Trantor Inc.
- Mayank Kumar, Machine Learning Engineer, Trantor Inc.
2. Agenda
• Main difference between Machine Learning and Software Development
• Difference b/w AI, Machine Learning (ML) and Deep Learning (DL)
• Why Now?
• Hype vs. Reality
• Difference b/w Supervised and Unsupervised ML
• Stages involved in an end-to-end ML project
• Technology stack commonly used in ML
• Broad categories (across domains) of use-cases in ML
• Trends in ML
• Challenges involved
• AI and Ethics?
• PoCs/Demos
• How/Where to get started?
• How much Maths is required?
• Quiz
17. Trends in ML: Transfer Learning
Source: https://medium.com/data-science-101/transfer-learning-57ce3b98650
18. Trends in ML:
• AutoML
• Transfer Learning
❑ Computer Vision (CV)
❑ Natural Language Processing (NLP)
• Cloud
Challenges involved in ML:
• Getting access to data
• Data annotation/labelling (GIGO)
• Iterative process
• Reproducibility
• "Black-box" problem
• Overfitting
23. AI and Ethics:
❖Data privacy
❖Surveillance
❖Post-truth politics
o GANs
❖Robots taking over?
❖Mass unemployment?
o Transition
o Full automation?
o Combo e.g. radiology
Link for GANs: https://www.youtube.com/watch?v=-cOYwZ2XcAc
40. 1. Most of the success stories (till date) in machine learning, have come from:
a) Supervised learning
b) Unsupervised learning
c) Reinforcement learning
d) None of the above
2. Which of the following, are true regarding AutoML?:
a) Adanet, TPOT, Featuretools and Autoweka are all tools for Auto ML.
b) It makes Data scientists’ job, redundant.
c) It will save more than half the time usually taken in an ML project.
d) All of the above.
3. Which of the following, is NOT an example of supervised learning?
a) Linear regression
b) Classification
c) Clustering
d) All of the above
41. 4. Most of the popular deep learning frameworks are based on:
a) R
b) Python
c) Julia
d) Scala
5. Pick the odd one out:
a) matplotlib
b) seaborn
c) plotly
d) beautiful soup
6. Classify the following statements as either True or False:
a) AI can do just about anything a human can do.
b) Model training is usually the most time-consuming part of an ML project.
c) ML is capable of dealing with new types of data too.
d) Combining different algorithms will improve the interpretability of the ML model.
42. 7. Identify the correct statement:
a) There is no free access to GPUs available through internet.
b) ML is something very new and recent.
c) In every use case of ML, it performs better than humans.
d) Most of the data out there, has been generated rather recently.
8. Which of the following statements are True?: (may be more than 1)
a) Transfer learning is working fine for CV, but not for NLP.
b) Neural networks are based on functioning of the human brain.
c) DL does not face the problem of overfitting.
d) DL is only a subset of ML.
9. Pick the odd one out:
a) Defining/Understanding the business problem
b) Model selection and training
c) Data procurement/ingestion
d) Monitoring of model predictions
43. 10. Which of the following, is NOT the best way to describe reasons for
democratization of AI?
a) More data
b) Open sourcing of algorithms/code
c) Reliable and faster connectivity
d) GPUs have become dirt cheap.