Created
September 8, 2021 23:07
-
-
Save luxzia/a9e8c4d284f0024ec8e5e33f8f1fe4bc to your computer and use it in GitHub Desktop.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
question | example XAI techniques | explanations description | |
---|---|---|---|
how (global model-wide) | ProfWeight or BRCG or GLRM | 1. Describe general model logic as feature impact or rules or decision-tree 2. if a user is only interested in a high-level view describe what are the top features or rules considered. | |
why | LIME or SHAP or ProtoDash | 1. describe what the key features of the inquired instance determine the model's prediction of it. 2. Show similar exampels with the same predicted outcome to justify the model's prediction | |
why not (a different prediction) | CEM or ProtoDash (on alternative prediction) | 1. describe what changes are required for the instance to get the alternative prediction and or what features of the instance guarantee the current prediction. 2. show prototypical examples that had the alternate outcomes. | |
how to change to be that (a different prediction) | CEM | higlight feature that if changed (by increasing or decreasing or making absent or making present) could alter the prediction | |
how to remain to be this (the current prediction) | CEM | describe what feature or feature range could guarantee the same prediction | |
what if | use feature tools such as sk-learn's PDP and ICE plots | show how the prediction changes corresponding to the inquired change | |
performance | precision or recall or accuracy or F1 or AUC. Quantify uncertainty using IBM Uncertainty Quantification 360. | 1. Provide performance metrics of the model. 2. Show uncertainty information for each prediction. 3. Describe potential strengths and limitations of the model | |
data | see examples in IBM FactSheets 360 | document comprehensive information abou the training data such as including the source or provenance or type or size or coverage of population or potential biases | |
output | see examples in IBM FactSheets 360 | 1. describe the scope of output or system functions. 2. suggest how the output should be used for downstream tasks or user workflow |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment