Machine Learning with Limited Labeled Data 4/3/19
5 likes3,661 views
The document discusses methodologies for selecting labeled data in machine learning, particularly under conditions of limited labeled data. It introduces concepts such as random sampling and uncertainty margin sampling, along with techniques for estimating distances from decision boundaries using adversarial perturbations. The document also acknowledges contributions from experts affiliated with the University of Chicago and Explosion AI.
Machine Learning with Limited Labeled Data 4/3/19
- 1. Confidential – Restricted LEARNING with LIMITED LABELED DATA Shioulin Sam and Nisha Muktewar Sanjay Krishnan (University of Chicago), Ines Montani (Explosion AI)
- 10. Confidential – Restricted 10
- 11. Confidential – Restricted 11
- 12. Confidential – Restricted 12
- 13. Confidential – Restricted 13
- 14. Confidential – Restricted 14
- 15. Confidential – Restricted 15
- 16. Confidential – Restricted 16
- 17. Confidential – Restricted Choosing what to label
- 18. Confidential – Restricted 18 Random Sampling
- 19. Confidential – Restricted 19 Uncertainty - Margin Sampling
- 20. Confidential – Restricted 20 Uncertainty - Margin Sampling
- 21. Confidential – Restricted Extending to deep learning
- 22. Confidential – Restricted 22 Distance from the decision boundary Computing the distance between the datapoint and the closest neighbor of a different class
- 23. Confidential – Restricted 23 Distance from the decision boundary Using adversarial perturbations to estimate the distance Image from https://openai.com/blog/adversarial-example-research/
- 24. Confidential – Restricted 24 Distance from the decision boundary Computing the distance between the datapoint and the decision boundary using perturbation magnitude
- 26. Confidential – Restricted 26
- 27. Confidential – Restricted 27
- 28. Confidential – Restricted 28
- 29. Confidential – Restricted 29
- 30. Confidential – Restricted 30 Ines Montani Founder of Explosion AI
- 31. Confidential – Restricted 31 Sanjay Krishnan Assistant Professor of Computer Science, University of Chicago
- 32. Confidential – Restricted 32 Q & A
- 33. Confidential – Restricted THANK YOU
- 34. Confidential – Restricted 34
- 35. Confidential – Restricted 35
- 36. Confidential – Restricted 36