On a random surfing of web I came across this paper Leveraging Uncertainties in Softmax Decision-Making Models for Low-Power IoT Devices by Chiwoo Cho et.al. In this paper the author proposed using Jain’s Fairness Index(JFI) to compute uncertanity in a deep learning model in IoT devices.
Instead of computing uncertanity, I thought of adding JFI as a suppliemtary criterion with our loss function to improve our model training. I have trained a model to classify iris dataset with and without Jain’s Fairness Index. There is a slight improvement in the model prediction when trained with Jain's Fairness Index.
The code for iris dataset training has been taken from here and I have only added jains_fairness_index()
funciton to it. I tested this idea with this dataset alone. I have added the seed