https://github.com/ayong8/FairSight

A visual analytic system for fair data-driven decision making
https://github.com/ayong8/FairSight

Last synced: 2 months ago
JSON representation

A visual analytic system for fair data-driven decision making

• Host: GitHub
• URL: https://github.com/ayong8/FairSight
• Owner: ayong8
• License: mit
• Created: 2018-06-29T17:27:51.000Z (about 6 years ago)
• Default Branch: master
• Last Pushed: 2022-12-10T16:55:12.000Z (over 1 year ago)
• Last Synced: 2024-01-27T14:32:27.929Z (5 months ago)
• Language: JavaScript
• Homepage:
• Size: 1.96 MB
• Stars: 22
• Watchers: 3
• Forks: 2
• Open Issues: 31
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Lists

• awesome-fairness-in-ai - FairSight: Visual Analytics for Fairness in Decision Making

README

        


	

	




FairSight is a viable fair decision making system to assist decision makers in achieving fair decision making through the machine learning workflow.


### Reference

Ahn, Y., & Lin, Y. R. (2019). Fairsight: Visual analytics for fairness in decision making. IEEE transactions on visualization and computer graphics (TVCG), 26(1), 1086-1095.

### Specification

- React: Frontend framework for rendering and communicating with data

- django: Python-based backend framework for serving API of data and running machine learning work

- scss: The stylesheet grammar for more flexible structure

- d3.js: Javascript-based visualization library

### FairDM



	

	




FairSight is developed on top of FairDM, a general fair decision making framework. Our framework is a model-agnostic framework with its goal to provide a fairness pipeline to guide the examination of fairness at each step (from input to output) in the workflow.


### System



	

	


	




(a) Generator: The workflow of FairSight starts with setting up inputs including the sensitive attribute and protected group. 


(b) Ranking View: After running a model, the ranking outcome and measures are represented. 


(c) Global Inspection View: Visualizes the two spaces and the mapping process of Individual and Group fairness provided in the separate tap. 


(d) Local Inspection View: When an individual is hovered, Local Inspection

View provides the instance- and group-level exploration. 


(e) Feature Inspection View: Users can investigate the feature distortion and feature perturbation to identify features as the possible source of bias. 


(f) Ranking List View: All generated ranking outcomes are listed and plotted.