cakiki / ml-fairness-validity

Exploring issues of fairness and validity in machine learning pipelines

 ______________________  +------- Wednesday, 22 May 2024 -------+ Christopher Akiki                
< Fortune is on hiatus > |                                      | ┣━━ Interests                    
 ----------------------  | Hello, friend.                       | ┃   ┣━━ My cat                   
       \ (__)            |                                      | ┃   ┣━━ Representation Learning  
         (oo)            | This auto-generated message panel    | ┃   ┣━━ Language Generation      
   /------\/             | was brought to you by the cowsay     | ┃   ┣━━ Text Mining              
  / |    ||              | apt, fortune and Rich.               | ┃   ┣━━ Dataset Creation         
 *  /\---/\              |                                      | ┃   ┗━━ TODO                     
    ~~   ~~              | Follow me on twitter: @christopher   | ┣━━ Past Lives                   
                         |                                      | ┃   ┣━━ Sociocultural antropology
                         +--------------------------------------+ ┃   ┗━━ Network Engineering      
                                                                  ┣━━ Current Location             
                                                                  ┃   ┗━━ Leipzig, Germany         
                                                                  ┗━━ Previous Locations           
                                                                      ┣━━ Durham, England          
                                                                      ┗━━ Zouk Mikael, Lebanon     

ml-fairness-validity's Issues

make lstm ready for slides

Conduct PCA to extract features & prepare for slides

Descriptives each dataset (plotted)

make random forest ready for slides

Finding join table & info for NUTS coding

Look at travel data and prepare

https://www-genesis.destatis.de/genesis//online?operation=table&code=45412-0014&bypass=true&levelindex=0&levelid=1598945781726#abreadcrumb

prepare pictures for slides

• picture with logo of google, facebook, etc.
• stop picture
• nice graphic for talking 30 sec. about corona (heat map?)

Create table with most interesting variables

Data introduction for slides

which data is nice to show so audience gets an impression of the structure?
e.g. explain percentage from baseline, meaning of tile etc

Prepare motivaton slides with content

Descriptives for each datasets (tables)

Compute fairness metrics on dataset

https://aif360.readthedocs.io/en/latest/modules/metrics.html

(https://github.com/IBM/AIF360)