Machine Learning Explainability
This one-day training course will help prevent your models from being a black box. It will provide you with a toolbox of Machine-Learning explainability techniques that you can use to explain your models to technical and non-technical people alike. You will also learn when, how, and why you should use the different techniques, and their drawbacks.
Looking to upskill your team(s) or organization?
Surat will gladly help you further with custom training solutions.
Surat Wahal
Training Advisor
What will you learn?
After the training, you will be able to:
Explain the use cases for model explainability
Evaluate when model explainability is not enough (correlation vs. causality, fairness)
Categorize the used methods into sensitivity vs. impact as well as explaining single predictions (local explainability) vs. multiple predictions (global explainability)
Apply the explainability methods with the provided Python packages
Summarize the advantages and disadvantages for each method
Evaluate whether a method is appropriate for the business use case
Program
- Introduction to Machine Learning interpretability/explainability
- Example-based explanations
- The inherent interpretability of ML models
- Ceteris Paribus plots
- Break-down Plots for Additive Attributions
- (Model-specific & Permutation) Feature importance
- Partial dependence plots
- Shapley and SHAP values
This training is for you if:
You are familiar with Data Science and Machine Learning concepts and can comfortably program in Python
You can already build and train machine learning models using Scikit-Learn
You want to better understand how your Machine Learning models make their predictions
You understand that your use case could fall under the “high risk AI applications” category of the EU AI act and explaining your model will be a must-have in the future.
This training is not for you if:
You have never programmed in Python before
You are unfamiliar with core Machine Learning concepts, such as train-test splits, model training and measuring model performance with metrics
You are only interested in achieving the best performance with your models and not how they achieved it
What else should I know?
After registering for this training, you will receive a confirmation email with practical information. A week before the training, the trainer will get in touch to ask you about any requirements you may have and any pre-course tasks you will need to do.
See you soon!
Course Information
All literature and course materials are included in the price
Information on the software and tooling will be shared before the start date
This course requires a laptop
Online courses are delivered via Zoom or Microsoft Teams
