You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
This repository was archived by the owner on May 9, 2024. It is now read-only.
Project Description: Min-max fairness is a natural and desirable notion of subgroup fairness. The goal of this project is to develop open source implementations of recent research into label efficient active sampling algorithms for achieving min-max fairness. Furthermore the project will involve benchmarking the algorithms against baselines, demonstrating fairness benefits on several publicly available datasets and exploring the use of accelerated gradient descent based techniques for making the algorithms more efficient.
Towards the above goal we encourage participants to first complete the following starter tasks.
Note: The following tasks are open for multiple submissions. Anyone interested in the project is encouraged so solve the tasks below and submit their work.
Starter Task 1: Train a one layer fully connected network in tensorflow to solve the binary classification problem of income prediction for the UCI Adult Dataset. Choose a reasonable batch size (say 32 or 64) and a reasonable number of epochs (say 50). Plot the test accuracy as a function of the number of processed batches. Also plot, as a function of the number of processed batches, the performance gap in the model accuracy for points with sex=Male vs. points with sex=Female.
Starter Task 2: Partition the training set into two subsets S1 and S2 based on sex. Retrain the network with the same settings as in Task 1. However this time alternate batch selection between S1 and S2. In other words if the first batch of 32 or 64 examples is chosen randomly from S1 then the next batch is chosen from S2..and so on. Report the same metrics as in Task 1. Summarize your results and suggest one or two ways to reduce the performance gap in the model accuracy for points with sex=Male vs. points with sex=Female.
Project Description: Min-max fairness is a natural and desirable notion of subgroup fairness. The goal of this project is to develop open source implementations of recent research into label efficient active sampling algorithms for achieving min-max fairness. Furthermore the project will involve benchmarking the algorithms against baselines, demonstrating fairness benefits on several publicly available datasets and exploring the use of accelerated gradient descent based techniques for making the algorithms more efficient.
Towards the above goal we encourage participants to first complete the following starter tasks.
Note: The following tasks are open for multiple submissions. Anyone interested in the project is encouraged so solve the tasks below and submit their work.
Starter Task 1: Train a one layer fully connected network in tensorflow to solve the binary classification problem of income prediction for the UCI Adult Dataset. Choose a reasonable batch size (say 32 or 64) and a reasonable number of epochs (say 50). Plot the test accuracy as a function of the number of processed batches. Also plot, as a function of the number of processed batches, the performance gap in the model accuracy for points with sex=Male vs. points with sex=Female.
Starter Task 2: Partition the training set into two subsets S1 and S2 based on sex. Retrain the network with the same settings as in Task 1. However this time alternate batch selection between S1 and S2. In other words if the first batch of 32 or 64 examples is chosen randomly from S1 then the next batch is chosen from S2..and so on. Report the same metrics as in Task 1. Summarize your results and suggest one or two ways to reduce the performance gap in the model accuracy for points with sex=Male vs. points with sex=Female.