Fairness-aware genetic-algorithm-based few-shot classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Artificial intelligence (AI)-assisted decision-making is prevalent, but biased data leads to unfair outcomes.
• Computational methods are essential to mitigate inequities in algorithmic decision-making processes.

Purpose of the Study:

• To present a novel framework for few-shot classification that integrates fair feature selection and fair meta-learning.
• To address and reduce unfairness stemming from biased data in AI-driven decisions.

Main Methods:

• A three-part framework: pre-processing for feature pool generation, a fair genetic algorithm (FairGA) for feature filtering, and fair few-shot (FairFS) for classification.
• FairGA utilizes word presence/absence as gene expression for fairness clustering and key feature identification.
• FairFS performs representation and fairness constraint classification, employing a combinatorial loss function for fairness and hard samples.

Main Results:

• The proposed method demonstrates strong competitive performance across three public benchmarks.
• The framework effectively integrates feature selection and meta-learning to enhance fairness in few-shot classification.

Conclusions:

• The developed framework offers a viable computational solution for achieving fairer algorithmic decision-making.
• This approach contributes to the development of more equitable AI systems by tackling data bias in classification tasks.