fairness

The paper shows that unequal single-token support for names in large language models leads to biased concept accessibility across demographics, and introduces NameTrace to measure lexical comparability.

The study systematically assesses algorithmic fairness in machine learning models for predicting treatment retention in medication for opioid use disorder, finding performance gaps across patient subgroups and evaluating bias mitigation techniques with trade-offs.

FairLMs is a Python library designed to streamline fairness research in language models by providing metrics, mitigation methods, and diagnostics with explicit capability declarations.

This paper proposes Fed-Equilibrium, a federated learning framework that balances robustness and fairness in clinical networks using topological Pareto control to ensure minority nodes achieve convergence comparable to dominant hubs.

The paper introduces a bias depth score to distinguish stable 'deep biases' from prompt-dependent 'shallow biases' in large language models, showing that deep biases are more persistent and harder to remove.

This paper introduces SCOPED-Hiring, a process-aware fairness diagnosis pipeline for LLM-based multi-agent hiring systems that uncovers hidden biases in decision trajectories and enables targeted repairs.

This paper proposes new bias testing methodologies for LLMs and VLMs in autonomous vehicles, showing that these models inherit human biases in pedestrian yielding decisions based on attributes like gender, ethnicity, and age.

This paper introduces a token-cost ledger to analyze the multilingual tokenization tax, decomposing it into removable and intrinsic components, showing that much excess token cost for non-English text is removable through improved tokenization codes.

RL-FAT is a reinforcement learning framework for fair adversarial training that improves robustness while reducing class-wise disparities. Experiments demonstrate competitive accuracy and better fairness compared to standard methods.

This paper proposes a unified framework for decentralized learning under communication constraints, introducing the DMFL-SQ algorithm that combines graph-based personalization, agnostic fairness, and compressed communication to achieve reduced communication while maintaining predictive performance and improving fairness across clients.

This paper proposes Subgraph Filtering for Fair Graph Neural Networks (SF-GNN), a framework that mitigates structural bias in GNNs by filtering edges to improve fairness-accuracy trade-offs.

This study examines how large language models judge urban safety based on neighborhood names versus coordinates, revealing that names carry both crime signal and demographic bias, with bias scaling with geographic knowledge.

This paper presents an interpretable feature-LLM hybrid system for automated L2 English speaking assessment that outperforms individual human raters and shows that pause encoding does not significantly impact LLM fluency scores.

Voice Arena and Hugging Face have added Hindi and Indian English to the Open ASR Leaderboard, introducing new evaluation sets designed to capture diverse speaker attributes and address biases in automatic speech recognition.

This paper introduces an adaptive triggering framework for bias correction in LLM chain-of-thought reasoning, using online change-point detection to optimize intervention timing with white-box and black-box signals, improving accuracy and reducing interventions.

This paper proposes Counterfactual Ensemble Decoding (CED) to mitigate social biases in large vision-language models by constructing multi-group counterfactual perspectives and integrating them during decoding, achieving substantial bias reduction while preserving model capabilities.

Redox OS has implemented the EEVDF scheduler, replacing DWRR, resulting in significant improvements in fairness, context switch time, and throughput.

The article discusses recent healthcare studies revealing that AI models inherit human biases, with specific examples from research by Zack et al. (2024) and Omar et al. (2025). It invites readers to share their experiences with AI bias in models like ChatGPT and Claude.

This paper proposes FedCurv-DR, a lightweight federated continual learning method that reduces forgetting and balances performance, fairness, and energy efficiency for sustainable AI in cultural heritage applications.

The paper proposes an evaluation framework for cross-lingual fairness in language model watermarking, revealing that disparities are structural to language typology rather than idiosyncratic to specific languages.