Adaptive Sensitive Reweighting

Description

Adaptive Sensitive Reweighting dynamically adjusts the importance of training examples during model training based on real-time performance across different demographic groups. Unlike traditional static reweighting that fixes weights at the start, this technique continuously monitors fairness metrics and automatically increases the weight of examples from underperforming groups whilst decreasing weights for overrepresented groups. The adaptive mechanism prevents models from perpetuating historical biases by ensuring balanced learning across all demographics throughout the training process.

Example Use Cases

Fairness

Training speech recognition systems that adapt weights during training to ensure consistent accuracy across different accents, dialects, and linguistic backgrounds, preventing models from favouring dominant accent groups in the training data.

Developing hiring algorithms that dynamically adjust training example weights to maintain consistent evaluation performance across demographic groups, ensuring the model doesn't learn to favour candidates from overrepresented backgrounds.

Reliability

Building medical diagnostic models that adaptively reweight patient examples during training to ensure reliable performance across different age groups, ethnicities, and socioeconomic backgrounds, preventing healthcare disparities.

Limitations

Training instability can occur when adaptive weight adjustments cause oscillations between demographic groups, potentially preventing convergence if reweighting parameters are not carefully tuned.
Computational overhead increases significantly due to continuous monitoring of fairness metrics across groups during training, requiring additional memory and processing time.
Risk of overfitting to specific demographic subgroups if the adaptation mechanism becomes too aggressive in correcting for observed performance disparities during training.
Requires careful hyperparameter tuning for adaptation rates and fairness thresholds, making the technique sensitive to configuration choices that may not generalise across different datasets.
May inadvertently harm overall model performance if the reweighting process prioritises fairness at the expense of learning important patterns that benefit all groups.

Resources

Research Papers

Adaptive Sensitive Reweighting to Mitigate Bias in Fairness-aware Classification

Emmanouil Krasanakis et al.•Jan 1, 2018

Causal Fairness-Guided Dataset Reweighting using Neural Networks

Xuan Zhao et al.•Nov 17, 2023

The importance of achieving fairness in machine learning models cannot be overstated. Recent research has pointed out that fairness should be examined from a causal perspective, and several fairness notions based on the on Pearl's causal framework have been proposed. In this paper, we construct a reweighting scheme of datasets to address causal fairness. Our approach aims at mitigating bias by considering the causal relationships among variables and incorporating them into the reweighting process. The proposed method adopts two neural networks, whose structures are intentionally used to reflect the structures of a causal graph and of an interventional graph. The two neural networks can approximate the causal model of the data, and the causal model of interventions. Furthermore, reweighting guided by a discriminator is applied to achieve various fairness notions. Experiments on real-world datasets show that our method can achieve causal fairness on the data while remaining close to the original data for downstream tasks.

Software Packages

AIF360

Aug 22, 2018

A comprehensive set of fairness metrics for datasets and machine learning models, explanations for these metrics, and algorithms to mitigate bias in datasets and models.

fairlearn

May 15, 2018

A Python package to assess and improve fairness of machine learning models.

Related Techniques

Name	Description	Assurance Goals
Fair Adversarial Networks	An in-processing fairness technique that employs adversarial training with dual neural networks to learn fair representations. The method consists of a predictor network that learns the main task whilst an adversarial discriminator network simultaneously attempts to predict sensitive attributes from the predictor's hidden representations. Through this adversarial min-max game, the predictor is incentivised to learn features that are informative for the task but statistically independent of protected attributes, effectively removing bias at the representation level in deep learning models.	Fairness Transparency Reliability
Meta Fair Classifier	An in-processing fairness technique that employs meta-learning to modify any existing classifier for optimising fairness metrics whilst maintaining predictive performance. The method learns how to adjust model parameters or decision boundaries to satisfy fairness constraints such as demographic parity or equalised odds through iterative optimisation. This approach is particularly valuable when retrofitting fairness to pre-trained models that perform well but exhibit bias, as it can incorporate fairness without requiring complete retraining from scratch.	Fairness Transparency Reliability
Exponentiated Gradient Reduction	An in-processing fairness technique based on Agarwal et al.'s reductions approach that transforms fair classification into a sequence of cost-sensitive classification problems. The method uses an exponentiated gradient algorithm to iteratively reweight training data, returning a randomised classifier that achieves the lowest empirical error whilst satisfying fairness constraints. This reduction-based framework provides theoretical guarantees about both accuracy and constraint violation, making it suitable for various fairness criteria including demographic parity and equalised odds.	Fairness Transparency Reliability
Fair Transfer Learning	An in-processing fairness technique that adapts pre-trained models from one domain to another whilst explicitly preserving fairness constraints across different contexts or populations. The method addresses the challenge that fairness properties may not transfer when adapting models to new domains with different demographic compositions or data distributions. Fair transfer learning typically involves constraint-aware fine-tuning, domain adaptation techniques, or adversarial training that maintains equitable performance across groups in the target domain, ensuring that bias mitigation efforts carry over from source to target domains.	Fairness Transparency Reliability