Recent research on bias mitigation in machine learning focuses on developing innovative strategies to address systemic biases in large language models (LLMs) and vision-language models (VLMs). Notably, new methodologies like diffusion-based style transfer are being employed to generate synthetic data that enhances representation in underrepresented demographics, particularly in mental health contexts. Simultaneously, frameworks combining category-theoretic transformations with retrieval-augmented generation are being proposed to structurally debias LLMs while maintaining semantic integrity. Other approaches aim to extract bias-free subnetworks from conventional models without additional data or retraining, enhancing computational efficiency. Additionally, addressing framing effects has emerged as a critical area, with methods designed to ensure consistent model responses across varied prompt expressions. These advancements not only aim to improve fairness in AI outputs but also have significant implications for applications in sensitive domains, such as healthcare and social media, where biased outputs can perpetuate harmful stereotypes and misinformation.
Synthetic data offers a promising solution for mitigating data scarcity and demographic bias in mental health analysis, yet existing approaches largely rely on pretrained large language models (LLMs),...
Biases in large language models (LLMs) often manifest as systematic distortions in associations between demographic attributes and professional or social roles, reinforcing harmful stereotypes across ...
The issue of algorithmic biases in deep learning has led to the development of various debiasing techniques, many of which perform complex training procedures or dataset manipulation. However, an intr...
As large language models (LLMs) are increasingly deployed in real-world applications, ensuring their fair responses across demographics has become crucial. Despite many efforts, an ongoing challenge i...
Large language models (LLMs) have demonstrated impressive capabilities across a wide range of natural language processing tasks. However, their outputs often exhibit social biases, raising fairness co...
Vision-Language Models (VLMs) are known to inherit and amplify societal biases from their web-scale training data with Indian being particularly misrepresented. Existing fairness-aware datasets have s...
Chronological age predictors often fail to achieve out-of-distribution (OOD) gen- eralization due to exogenous attributes such as race, gender, or tissue. Learning an invariant representation with res...