Slides for my talk "Machine Learning by examples" in the Introduction to Computing course at Texas A&M University (TAMU), Spring 2024

1. Machine Learning by
Examples
Marcus Botacin

2. Machine Learning Tasks

3. Separate in groups

4. Classification

5. Separate in groups

6. Clustering

7. ML Features

8. ML: Feature Extraction and Representation
Source:
https://manningbooks.medium.com/the-computer-vision-pipeline-part-4-feature-extra
ction-6343ef063588

9. ML Pipelines
Source: https://www.nature.com/articles/srep06367

10. How do the models learn?

11. Is this the best way to separate these points?

12. Is this the best way to separate these points?

13. Is this the best way to separate these points?

14. How complex models can be?

15. Non-Linear Models
Source: Stanford NLP Group

16. Is it as easy as keeping extracting features?

17. What is a cat?

18. Is this a cat?

19. A cat has no wings

20. Is this a cat?

21. A cat has no wings and 4 legs

22. Is this a cat?

23. A cat has no wings, 4 legs, and it is small

24. Is this a cat?

25. What is a cat?

27. Deep Learning

28. ML vs. DL: Concepts
Source: https://blog.dataiku.com/when-and-when-not-to-use-deep-learning

29. DL Application: Face Recognition
Source: https://thehackernews.com/2021/11/facebook-to-shut-down-facial.html
Source:
https://www.thehansindia.com/technology/tech-news/facebook-to-shut-down-facial-re
cognition-feature-713722

30. And now we are good, right?

31. Adversarial Machine Learning
Source:
https://towardsdatascience.com/breaking-neural-networks-with-adversarial-attack
s-f4290a9a45aa

32. AML: Malware Detector Evasion
Source:
https://marcusbotacin.github.io/publication/2019-01-01-paper-evasion-number-12

33. But it’s OK in controlled environments, right?

34. DL: Lack of Explainability
Source:
https://www.consumerfinance.gov/about-us/newsroom/cfpb-acts-to-protect-the-pu
blic-from-black-box-credit-models-using-complex-algorithms/

35. But it’s just to limit it to non-sensitive
applications, right?

36. DL: Lack of Explainability
Source:
https://www.forbes.com/sites/kashmirhill/2012/02/16/how-target-figured-out-a-tee
n-girl-was-pregnant-before-her-father-did/?sh=979604166686

37. Natural Language Processing (NLP), Large
Language Models (LLMs), and ChatGPT

38. Complete the statements…
1. Happy New …
2. Merry …

39. How does it work?

40. NLP: Tokenization
Source: https://www.freecodecamp.org/news/evolution-of-tokenization/

41. NLP: Text Generation
Source:
https://www.sitepen.com/blog/exploring-the-creative-possibilities-of-markov-chain
s-for-text-generation

42. Nothing might go wrong, right?

43. ChatGPT: Automatic Attacks
Source:
https://marcusbotacin.github.io/publication/2023-05-01-paper-gpt-number-27

44. But this does not happen in reality, right?
Right?

45. FraudGPT: Malicious LLMs
Source:
https://www.trustwave.com/en-us/resources/blogs/spiderlabs-blog/wormgpt-and-fr
audgpt-the-rise-of-malicious-llms/

46. How to solve these new problems?

47. Questions?
A long road Ahead!