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.

Interpretable Machine Learning

169 views

Published on

Interpretierbarkeit von ML-Modellen hat die Zielsetzung, die Ursachen einer Prognose offenzulegen und eine daraus abgeleitete Entscheidung für einen Menschen nachvollziehbar zu erklären. Durch die Nachvollziehbarkeit von Prognosen lässt sich beispielsweise sicherstellen, dass deren Herleitung konsistent zum Domänenwissen eines Experten ist. Auch ein unfairer Bias lässt sich durch die Erklärung aussagekräftiger Beispiele identifizieren.

Prognosemodelle lassen sich grob in intrinsisch interpretierbare Modelle und nicht-interpretierbare (auch Blackbox-) Modelle unterscheiden. Intrinsisch interpretierbare Modelle sind dafür bekannt, dass sie für einen Menschen leicht nachvollziehbar sind. Ein typisches Beispiel für ein solches Modell ist der Entscheidungsbaum, dessen regelbasierter Entscheidungsprozess intuitiv und leicht zugänglich ist. Im Gegensatz dazu gelten Neuronale Netze als Blackbox-Modelle, deren Prognosen durch die komplexe Netzstruktur schwer nachvollziehbar sind.

In diesem Talk erläuterte Marcel Spitzer das Konzept von Interpretierbarkeit im Kontext von Machine Learning und stellte gängige Verfahren zur Interpretation von Modellen vor. Besonderen Fokus legte er dabei auf modellunabhängige Verfahren, die sich auch auf prognosestarke Blackbox-Modelle anwenden lassen.

Event: M3 Minds Mastering Machines
Speaker: Marcel Spitzer

Blog-Artikel: https://www.inovex.de/blog/machine-learning-interpretability/

Mehr Tech-Vorträge: inovex.de/vortraege
Mehr Tech-Artikel: inovex.de/blog

Published in: Software
  • DOWNLOAD FULL BOOKS, INTO AVAILABLE FORMAT ......................................................................................................................... ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... 1.DOWNLOAD FULL. PDF EBOOK here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB Ebook here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... 1.DOWNLOAD FULL. doc Ebook here { https://tinyurl.com/y6a5rkg5 } ......................................................................................................................... ......................................................................................................................... ......................................................................................................................... .............. Browse by Genre Available eBooks ......................................................................................................................... Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, Cookbooks, Crime, Ebooks, Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult,
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • DOWNLOAD FULL eBOOK INTO AVAILABLE FORMAT ......................................................................................................................... ......................................................................................................................... 1.DOWNLOAD FULL. PDF eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. doc eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. PDF eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. EPUB eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... 1.DOWNLOAD FULL. doc eBook here { https://tinyurl.com/y3nhqquc } ......................................................................................................................... ......................................................................................................................... ......................................................................................................................... .............. Browse by Genre Available eBooks ......................................................................................................................... Art, Biography, Business, Chick Lit, Children's, Christian, Classics, Comics, Contemporary, CookeBOOK Crime, eeBOOK Fantasy, Fiction, Graphic Novels, Historical Fiction, History, Horror, Humor And Comedy, Manga, Memoir, Music, Mystery, Non Fiction, Paranormal, Philosophy, Poetry, Psychology, Religion, Romance, Science, Science Fiction, Self Help, Suspense, Spirituality, Sports, Thriller, Travel, Young Adult,
       Reply 
    Are you sure you want to  Yes  No
    Your message goes here
  • Be the first to like this

Interpretable Machine Learning

  1. 1. 2 ● Applied Mathematics, Data Science ● SW Engineering, Machine Learning ● Big Data, Hadoop, Spark mspitzer@inovex.de @mspitzer243
  2. 2. 3 Interpretation is the process of giving explanations to humans. ~ B. Kim, Google Brain, Interpretable Machine Learning (ICML 2017) https://people.csail.mit.edu/beenkim/papers/BeenK_FinaleDV_ICML2017_tutorial.pdf
  3. 3. 4 “Interpretability is the degree to which an observer can understand the cause of a decision.” ~ Miller T., 2017, Explanation in AI: Insights from the Social Sciences ● humans create decision systems ● humans are affected by decisions ● humans demand for explanations https://arxiv.org/pdf/1706.07269.pdf
  4. 4. 5 Bias towards Accuracy Do also consider asking yourself: ● Am I solving the right problem? ● How to make people trust my algorithm? ● Is there any bias? Are the training data representative? ● What is the impact in a real-world setting? https://bentoml.com/posts/2019-04-19-one-model/ “The machine learning community focuses too much on predictive performance. But machine learning models are always a small part of a complex system.” ~ C. Molnar, 2019, One Model to Rule Them All
  5. 5. 6https://arxiv.org/abs/1602.04938
  6. 6. 7https://arxiv.org/pdf/1606.03490.pdf The additional need for interpretability
  7. 7. 8https://arxiv.org/pdf/1606.03490.pdf The additional need for interpretability The decision process of a model should be consistent to the domain knowledge of an expert. In particular, it ... ● should not encode bias ● should not pick up random correlation ● should not use leaked information hard to capture in asingle quantity
  8. 8. 9 Use models that are intrinsically interpretable and known to be easy for humans to understand. Train a black box model and apply post-hoc interpretability techniques to provide explanations. 1 2
  9. 9. 10 Linear Regression Linearity makes model easy to interpret ● learned weights can be used to explain feature effects ● predictions can be decomposed into individual attributions ● confidence intervals express uncertainty
  10. 10. Linear Regression: Model Internals https://christophm.github.io/interpretable-ml-book/limo.html ● Coefficients indicate direction of influence ● Confidence intervals express significance and uncertainty ● Not comparable across features unless standardized
  11. 11. Linear Regression: Feature Effects https://christophm.github.io/interpretable-ml-book/limo.html ● Decompose predictions into individual feature attributions ● Scale-independent, hence comparable across features ● Enables us to draw conclusions about feature importance
  12. 12. https://christophm.github.io/interpretable-ml-book/limo.html Linear Regression: Feature Effects ● Decompose predictions into individual feature attributions ● Scale-independent, hence comparable across features ● Enables us to draw conclusions about feature importance ● Individual effects comparable to overall distribution
  13. 13. 14 Decision Trees Intuitive decision process makes model easy to interpret ● captures nonlinear dependencies and interactions ● model is simple and self-describing ● rule-based prediction feels natural for humans
  14. 14. 15 Decision Trees: Model Internals income > X? dept < Y? YESNO NO has larger impact than dept ● Tree structure lets us assess feature importance
  15. 15. 16 Decision Trees: Model Internals income > X? dept < Y? YESNO NO ● Tree structure lets us assess feature importance ● Feature interactions can be determined by following paths from root to leaf
  16. 16. 17 Decision Trees: Individual Explanations income > X? dept < Y? YESNO NO ● Tree structure lets us assess feature importance ● Feature interactions can be determined by following paths from root to leaf ● To explain a particular prediction, just take a look at the path from root to leaf
  17. 17. 18 Wrap Up: Desirable Properties Intrinsically interpretable models simplify answering: ● Which features are relevant? ● How do they influence predictions? ● How do features interact? ● How certain is a prediction? Both for the entire model as well as individual predictions
  18. 18. 19 Wrap Up: Desirable Properties Intrinsically interpretable models simplify answering: ● Which features are relevant? ● How do they influence predictions? ● How do features interact? ● How certain is a prediction? Both for the entire model as well as individual predictions How to answer these questions if models were black boxes?
  19. 19. 20 Use models that are intrinsically interpretable and known to be easy for humans to understand. Train a black box model and apply post-hoc interpretability techniques to provide explanations. 1 2
  20. 20. 21 Feature Shuffling https://amunategui.github.io/variable-importance-shuffler/ Train Model Apply Shuffle Evaluate Training Testing Shuffled Scored RMSE’=4.8 Evaluate Scored RMSE=3.2 X X’ For every Feature X: Repeat this process N times X’ = Perm(X) Degradation(RMSE,X) = 4.8 - 3.2 = 1.6
  21. 21. 22 Feature Shuffling https://christophm.github.io/interpretable-ml-book/feature-importance.html https://scikit-plot.readthedocs.io/en/stable/estimators.html#scikitplot.estimators.plot_feature_importances ● Estimates Feature importance by averaging degradation ● Tied to certain loss function ● Not applicable in high dimensional domains
  22. 22. 23 Individual Conditional Expectation (ICE) https://christophm.github.io/interpretable-ml-book/pdp.html Train Model Training Testing Overwrite Modified X Apply Scored X’ X’ = Const XFor every Feature X: Repeat this process using some constants CX ŷ
  23. 23. 24 Individual Conditional Expectation (ICE) https://christophm.github.io/interpretable-ml-book/pdp.html
  24. 24. 25 Individual Conditional Expectation (ICE) https://christophm.github.io/interpretable-ml-book/pdp.html
  25. 25. 26 Individual Conditional Expectation (ICE) https://christophm.github.io/interpretable-ml-book/pdp.html
  26. 26. 27 Global Surrogate Models https://www.oreilly.com/ideas/ideas-on-interpreting-machine-learning
  27. 27. ● Feeds original model with small variations of instance to be explained ● Sampled instances are weighted by proximity to the instance of interest ● Interpretable models are fit locally on observed outcome 28 Local Surrogate Models (LIME) https://christophm.github.io/interpretable-ml-book/lime.html
  28. 28. 29 Local Surrogate Models (LIME) https://www.oreilly.com/learning/introduction-to-local-interpretable-model-agnostic-explanations-lime
  29. 29. 30 Conclusion ● performance metrics are crucial for evaluation, but they lack explanations ● criteria like fairness and consistency are much harder if not impossible to quantify ● the problem with blackboxes is the lack of trust caused by their opaque nature ● transparency is key to achieving trust and acceptance in the mainstream
  30. 30. 31https://xkcd.com/1838/ Conclusion don’t end up like this!
  31. 31. ● Molnar C., 2018, Interpretable Machine Learning - A Guide for Making Black Box Models Explainable ● Gill N., Hall P., 2018, An Introduction to Machine Learning Interpretability ● Zhao Q., Hastie T., 2017, Causal Interpretations of Black-Box Models ● Kim B., Doshi-Velez F., 2017, Interpretable Machine Learning: The fuss, the concrete and the questions ● Ribeiro, M.T., Singh, S. and Guestrin, C., 2016, August. Why should i trust you? Explaining the predictions of any classifier 32 Resources
  32. 32. 33
  33. 33. Marcel Spitzer Big Data Scientist mspitzer@inovex.de inovex GmbH Schanzenstraße 6-20 Kupferhütte 1.13 51063 Köln

×