SLIDE 1
A black-box model An interpretable model
+ High predictive performance + interpretable
A hybrid of both?
- non-interpretable
- lower predictive performance
Poster #67
Partial Substitute tong-wang@uiowa.edu Poster #67 A black-box - - PowerPoint PPT Presentation
Partial Substitute tong-wang@uiowa.edu Poster #67 A black-box - - PowerPoint PPT Presentation
Tong Wang Gaining Free or Low-Cost University of Iowa Transparency with Interpretable Tippie College of Business Partial Substitute tong-wang@uiowa.edu Poster #67 A black-box model + High predictive performance - non-interpretable An
SLIDE 2
SLIDE 3
A black-box model An interpretable model
+ High predictive performance + interpretable
A hybrid of both?
- non-interpretable
- lower predictive performance
A key observation: there might exist a subspace where a black-box is overkill and a simple interpretable model can perform just as well as the black-box Poster #67
SLIDE 4
A black-box model An interpretable model
A hybrid of both? A key observation: there might exist a subspace where a black-box is overkill and a simple interpretable model can perform just as well as the black-box The proposed solution: to substitute the black-box model with an interpretable model, where there is no or low- cost of predictive performance
+ High predictive performance + interpretable
An effective trade-off Poster #67
SLIDE 5
A black-box model An interpretable model
A hybrid of both? A key observation: there might exist a subspace where a black-box is overkill and a simple interpretable model can perform just as well as the black-box
+
- Predicted by an
interpretable model Predicted by an interpretable model The proposed solution: to substitute the black-box model with an interpretable model, where there is no or low- cost of predictive performance
+ High predictive performance + interpretable
An effective trade-off Predicted by a black-box model Poster #67
SLIDE 6
A A hybrid rid pr predi dict ctive mod
- del
Define transparency of model: πΈπ
πΈ
Poster #67
SLIDE 7
A A hybrid rid pr predi dict ctive mod
- del
Define transparency of model: πΈπ
πΈ
Learning Objective
- Predictive performance
- Interpretability of π
π
- Transparency
Poster #67
SLIDE 8
A A hybrid rid pr predi dict ctive mod
- del
A Hybrid Rule Set Define transparency of model: πΈπ
πΈ
Learning Objective
- Predictive performance
- Interpretability of π
π
- Transparency
Poster #67 Poster #67
SLIDE 9
Model Training
Any pre-trained black box classifier
{ΰ· π§ππ}π=1
π
{π¦π}π=1
π
training data {(π¦π, π§π)}π=1
π
Input of the training algorithm Training algorithm and Stochastic Local Search based algorithm (see the paper for more details)
Predicted labels {ΰ· π§ππ}π=1
π
Poster #67
SLIDE 10
Evaluation: An efficient frontier that characterizes the trade-off between transparency and accuracy
Poster #67
SLIDE 11
Performance on Juvenile dataset
transparency accuracy Poster #67
SLIDE 12