Data-Algorithm-Architecture Co-Optimization for Fair Neural Networks on Skin Lesion Dataset

TL;DR

This work tackles fairness in dermatology AI where skin-tone bias skews model performance across groups. It proposes BiaslessNAS, a neural-architecture-search framework that co-optimizes data batching, fairness-aware training, and network architecture under a unified objective. The method uses a reinforcement-learning controller to search both batch-generation policies and architectures, guided by fairness $U(f_N',D)$ and accuracy $A(f_N',D)$ metrics. On a multi-source skin-lesion dataset, BiaslessNAS achieves substantial fairness gains with competitive or improved accuracy, demonstrating the value of architecture-aware fairness for medical AI.

Abstract

As Artificial Intelligence (AI) increasingly integrates into our daily lives, fairness has emerged as a critical concern, particularly in medical AI, where datasets often reflect inherent biases due to social factors like the underrepresentation of marginalized communities and socioeconomic barriers to data collection. Traditional approaches to mitigating these biases have focused on data augmentation and the development of fairness-aware training algorithms. However, this paper argues that the architecture of neural networks, a core component of Machine Learning (ML), plays a crucial role in ensuring fairness. We demonstrate that addressing fairness effectively requires a holistic approach that simultaneously considers data, algorithms, and architecture. Utilizing Automated ML (AutoML) technology, specifically Neural Architecture Search (NAS), we introduce a novel framework, BiaslessNAS, designed to achieve fair outcomes in analyzing skin lesion datasets. BiaslessNAS incorporates fairness considerations at every stage of the NAS process, leading to the identification of neural networks that are not only more accurate but also significantly fairer. Our experiments show that BiaslessNAS achieves a 2.55% increase in accuracy and a 65.50% improvement in fairness compared to traditional NAS methods, underscoring the importance of integrating fairness into neural network architecture for better outcomes in medical AI applications.

Figures (4)

• Figure 1: Bias issue behind training dataset and three fairness-related factors
• Figure 2: Overview of BiaslessNAS : ➀ controller: generating a reward $R$ and updating the recurrent neural network (RNN)-based controller; ➁ search space: sampling a set of hyperparameters based on the updated controller to obtain the batch composition of groups' data and a child network; ➂ fairness-aware trainer: on a validated dataset, training the identified child network on the generated batches; ➃ evaluator: generate the accuracy and unfairness score for the trained neural network $f_N^{\prime}$.
• Figure 3: Visualization of BiaslessNAS-Fair and BiaslessNAS-Acc, together with their performance on different fairness metrics
• Figure 4: Evaluation of fairness-aware trainer on the existing neural architectures