The Effect of Model Compression on Fairness in Facial Expression Recognition

TL;DR

This study investigates how model compression techniques affect the fairness and accuracy of facial expression recognition neural networks across two datasets, revealing that compression can reduce size with minimal accuracy loss but may impact fairness differently depending on the dataset.

Contribution

The paper extends prior research by analyzing the impact of various model compression methods on fairness in facial expression recognition, comparing results across two datasets.

Findings

Compression reduces model size significantly with minimal accuracy loss.

RAF-DB shows robustness to compression compared to CK+DB.

Compression amplifies biases in CK+DB but not in RAF-DB.

Abstract

Deep neural networks have proved hugely successful, achieving human-like performance on a variety of tasks. However, they are also computationally expensive, which has motivated the development of model compression techniques which reduce the resource consumption associated with deep learning models. Nevertheless, recent studies have suggested that model compression can have an adverse effect on algorithmic fairness, amplifying existing biases in machine learning models. With this project we aim to extend those studies to the context of facial expression recognition. To do that, we set up a neural network classifier to perform facial expression recognition and implement several model compression techniques on top of it. We then run experiments on two facial expression datasets, namely the Extended Cohn-Kanade Dataset (CK+DB) and the Real-World Affective Faces Database (RAF-DB), to examine the individual and combined effect that compression techniques have on the model size, accuracy and fairness. Our experimental results show that: (i) Compression and quantisation achieve significant reduction in model size with minimal impact on overall accuracy for both CK+DB and RAF-DB; (ii) in terms of model accuracy, the classifier trained and tested on RAF-DB seems more robust to compression compared to the CK+ DB; (iii) for RAF-DB, the different compression strategies do not seem to increase the gap in predictive performance across the sensitive attributes of gender, race and age which is in contrast with the results on the CK+DB, where compression seems to amplify existing biases for gender. We analyse the results and discuss the potential reasons for our findings.