Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.

Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant Professor of Computer Science, UC Irvine

3,482 views

Published on

Presented at #H2OWorld 2017 in Mountain View, CA.

Enjoy the video: https://youtu.be/TBJqgvXYhfo.

Learn more about H2O.ai: https://www.h2o.ai/.

Follow @h2oai: https://twitter.com/h2oai.

- - -

Abstract:
Machine learning is at the forefront of many recent advances in science and technology, enabled in part by the sophisticated models and algorithms that have been recently introduced. However, as a consequence of this complexity, machine learning essentially acts as a black-box as far as users are concerned, making it incredibly difficult to understand, predict, or "trust" their behavior. In this talk, I will describe our research on approaches that explain the predictions of ANY classifier in an interpretable and faithful manner.

Sameer's Bio:
Dr. Sameer Singh is an Assistant Professor of Computer Science at the University of California, Irvine. He is working on large-scale and interpretable machine learning applied to natural language processing. Sameer was a Postdoctoral Research Associate at the University of Washington and received his PhD from the University of Massachusetts, Amherst, during which he also worked at Microsoft Research, Google Research, and Yahoo! Labs on massive-scale machine learning. He was awarded the Adobe Research Data Science Faculty Award, was selected as a DARPA Riser, won the grand prize in the Yelp dataset challenge, and received the Yahoo! Key Scientific Challenges fellowship. Sameer has published extensively at top-tier machine learning and natural language processing conferences. (http://sameersingh.org)

Published in: Technology
  • Login to see the comments

Explaining Black-Box Machine Learning Predictions - Sameer Singh, Assistant Professor of Computer Science, UC Irvine

  1. 1. Explaining Black-Box Machine Learning Predictions Sameer Singh University of California, Irvine
  2. 2. Machine Learning is Everywhere…
  3. 3. Classification: Wolf or a Husky? Machine Learning Model Wolf!
  4. 4. Classification: Wolf or a Husky? Machine Learning Model Husky!
  5. 5. Only 1 mistake! Classification: Wolf or a Husky?
  6. 6. More Complex: Question Answering Is there a moustache in the picture? > Yes What is the moustache made of? > Banana
  7. 7. Essentially black-boxes! How can we trust the predictions are correct? How do we know they are not breaking regulations? How do we avoid “stupid mistakes”? Trust How can we understand and predict the behavior? Predict How do we improve it to prevent potential mistakes? Improve
  8. 8. Only 1 mistake! Classification: Wolf or a Husky? We’ve built a snow detector…
  9. 9. Visual Question Answering What is the moustache made of? > Banana What are the eyes made of? > Bananas What? > Banana What is? > Banana
  10. 10. Text Classification Why did this happen? From: Keith Richards Subject: Christianity is the answer NTTP-Posting-Host: x.x.com I think Christianity is the one true religion. If you’d like to know more, send me a note
  11. 11. Applying for a Loan Machine Learning I would like to apply for a loan. vvvvvvv vvvvvvv vvvvvvv vvvvvvv vvvvvvv Here is my information. vvvvvvv vvvvvvv vvvvvvv vvvvvvv vvvvvvv Sorry, your request has been denied Why? What were the reasons? Currently Cannot explain.. [0.25,-4.5,3.5,-10.4,…]
  12. 12. How did we get here? Big Data and Deep Learning
  13. 13. Simple Data X1 X2
  14. 14. Linear Classifiers X1 X2 You can interpret it… - Both have a positive effect - X1 > X2 10X1 + X2 - 5 > 0if: otherwise
  15. 15. Decision trees X1 X2 X1 > 0.5 X2 > 0.5 You can interpret it… - X2 is irrelevant if X1<0.5 - Otherwise X2 is enough
  16. 16. Looking at the structure How can we trust the predictions are correct? Trust How can we understand and predict the behavior? Predict How do we improve it to prevent potential mistakes? Improve Test whether the structure agrees with our intuitions. Structure tells us exactly what will happen on any data. Structure tells you where the error is, thus how to fix it.
  17. 17. Arrival of Big Data
  18. 18. Big Data: Applications of ML
  19. 19. Big Data: More Complexity X1 X2
  20. 20. http://www.mckinsey.com/industries/high-tech/our-insights/an-executives-guide-to-machine-learning
  21. 21. Big Data: More Dimensions Savings Income Profession Loan Amount Age Marital Status Past defaults Credit scores Recent defaults This easily goes to hundreds - Images: thousands - Text: tens of thousands - Video: millions - … and so on
  22. 22. X1 X2 Complex Surfaces Savings Income Profession Loan Amount Age Married Past defaults Credit scores Recent defaults … Lots of dimensions + Black-boxes!
  23. 23. Accuracy vs Interpretability Interpretability Accuracy 10X1 + X2 - 5 > 0 X1 > 0.5 X2 > 0.5 millions of weights, complex features Real-world use case Research on “interpretable models”
  24. 24. Deep Learning Interpretability Accuracy Real-world use case Research on “interpretable models” Focus on accuracy! Human-level
  25. 25. Looking at the structure How can we trust the predictions are correct? Trust How can we understand and predict the behavior? Predict How do we improve it to prevent potential mistakes? Improve Test whether the structure agrees with our intuitions. Structure tells us exactly what will happen on any data. Structure tells you where the error is, thus how to fix it.
  26. 26. Explaining Predictions The LIME Algorithm
  27. 27. Being Model-Agnostic… No assumptions about the internal structure… X1 > 0.5 X2 > 0.5f(x) Explain any existing, or future, model Data Decision
  28. 28. LIME: Explain Any Classifier! Interpretability Accuracy Real-world use case Make everything interpretable!
  29. 29. What an explanation looks like Why did this happen? From: Keith Richards Subject: Christianity is the answer NTTP-Posting-Host: x.x.com I think Christianity is the one true religion. If you’d like to know more, send me a note
  30. 30. Being Model-Agnostic… “Global” explanation is too complicated
  31. 31. Being Model-Agnostic… “Global” explanation is too complicated
  32. 32. Being Model-Agnostic… “Global” explanation is too complicated Explanation is an interpretable model, that is locally accurate
  33. 33. Google’s Object Detector P( ) = 0.21P( ) = 0.24P( ) = 0.32
  34. 34. Only 1 mistake! Classification: Wolf or a Husky?
  35. 35. Neural Network Explanations We’ve built a great snow detector…
  36. 36. Understanding Behavior We’ve built a great snow detector…
  37. 37. Comparing Classifiers Classifier 1 Classifier 2 Explanations? Look at Examples? Deploy and Check? “I have a gut feeling..” Accuracy? Change the model Different data Different parameters Different “features” …
  38. 38. Comparing Classifiers Original Image “Bad” Classifier “Good” Classifier
  39. 39. Explanation for a bad classifier From: Keith Richards Subject: Christianity is the answer NTTP-Posting-Host: x.x.com I think Christianity is the one true religion. If you’d like to know more, send me a note After looking at the explanation, we shouldn’t trust the model!
  40. 40. “Good” Explanation It seems to be picking up on more reasonable things.. good!
  41. 41. Recent Work Counter-examples and Counter-factuals
  42. 42. Understanding via Predicting Users “understand” a model if they can predict its behavior on unseen instances Precision is much more important than Coverage! Precision How accurate are the users guesses? If the users guess wrong, they don’t understand Coverage How often do the users make confident guesses? It’s okay not to be able to guess! It’s much better not to guess than to guess confidently, but be completely wrong!
  43. 43. Linear Explanations This movie is not bad. This movie is not very good. LIMELIME D D … D This director is always bad. This movie is not nice. This stuff is rather honest. This star is not bad. Problem 1: Where is the explanation good? Explanation is wrong in this region This explanation is a better approximation than the other one. Problem 2: What is the coverage? Explanation doesn’t apply here → Users will make mistakes!
  44. 44. Anchors: Precise Counter-factuals Anchor: ”not bad” → This movie is not bad. This audio is not bad. This novel is not bad. This footage is not bad. D(.|A) Positive This movie is not very good. Anchor: ”not good” → This poster is not ever good. This picture is not rarely good. This actor is not incredibly good. D(.|A) Negative anchor anchor LIMELIME D D An anchor is a sufficient condition Clear (and adaptive) coverage Probabilistic guarantee avoids human mistakes
  45. 45. Salary Prediction IF Education < High School Then Predict Salary < 50K Salary 71% 29% >$50K <$50K
  46. 46. Visual QA
  47. 47. Encoder/Decoder LSTMs
  48. 48. Encoder/Decoder LSTMs
  49. 49. Encoder/Decoder LSTMs
  50. 50. What’s a Good Explanation? We want to understand the models Compact description Lines, Decision Trees, Simple Rules, etc. When we read them, we imagine instances where they apply, and where they don’t Directly show useful examples? What examples describe the behavior? Closest Counter-example: How can we change this example to change the prediction?
  51. 51. Adversarial Examples Goodfellow et al, "Explaining and Harnessing Adversarial Examples", ICLR 2015. adversary predicted as "2"original MNIST digit "3" + .02 x = adversarial noise "inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed input results in the model outputting an incorrect answer with high confidence"
  52. 52. Adversarial Examples: Pros Advantages: ◦ Applicable to any gradient -based classifier ◦ Useful to evaluate the robustness of the model against adversaries ◦ Small perturbations often lead to imperceivable adversarial examples
  53. 53. Adversarial Examples: Cons Disadvantages: ◦ Examples are unnatural ◦ may not look anything you would naturally see in the "wild" ◦ Distance is not always meaningful ◦ E.g. color change or translation/rotation of an image ◦ Cannot be used for structured domains like text, code, etc.: ◦ E.g. replacing/removing words results in sentences that are not grammatical ◦ Do not provide insights into why the sample is an adversary ◦ How is the model working? ◦ How to fix the model?
  54. 54. Example: MNIST Digits
  55. 55. Example: Church vs Tower
  56. 56. Machine Translation Debug Google Translate, remotely!
  57. 57. Explanations are important! How can we trust the predictions are correct? Trust How can we understand and predict the behavior? Predict How do we improve it to prevent potential mistakes? Improve Model Agnostic Explanations
  58. 58. Thanks! sameer@uci.edu sameersingh.org Model Agnostic Explanations Work with Marco T. Ribeiro, Carlos Guestrin, Dheeru Dua, and Zhengli Zhao

×