Debiasing Text Safety Classifiers through a Fairness-Aware Ensemble (2024.emnlp-industry)
Copied to clipboard
| Challenge: | Increasing use of large language models (LLMs) require performant guardrails to ensure the safety of inputs and outputs . when these guardrail are trained on imbalanced data, they can learn the societal biases resulting from the model's performance. |
| Approach: | They propose a method for mitigating counterfactual fairness in closed-source text safety classifiers by using a debiasing regularizer and a threshold-agnostic metric. |
| Outcome: | The proposed method outperforms classifiers and acts as a debiasing regularizer . it uses threshold-agnostic metrics and Fair Data Reweighting (FDW) to assess the counterfactual fairness of a model . |
Similar Papers
Counterfactual Inference for Text Classification Debiasing (2021.acl-long)
Copied to clipboard
| Challenge: | Existing methods to capture unintended dataset biases are expensive and require elaborate balancing strategies. |
| Approach: | They propose a model-agnostic text classification debiasing framework which can effectively avoid employing data manipulations or designing balancing mechanisms. |
| Outcome: | The proposed framework can effectively avoid data manipulations or designing balancing mechanisms to capture unintended dataset biases. |
A Group Fairness Lens for Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods focusing on a few groups lack a comprehensive categorical perspective to evaluate LLMs’ potential biases and unfairness. |
| Approach: | They propose to evaluate LLM biases from a group fairness lens using a hierarchical schema characterizing diverse social groups. |
| Outcome: | The proposed method mitigates biases in LLMs from a group fairness lens and encapsulates target-attribute combinations across multiple dimensions. |
BiasFilter: An Inference-Time Debiasing Framework for Large Language Models (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods for debiasing large language models incur high human and computational costs and are limited in their effectiveness. |
| Approach: | They propose a model-agnostic, inference-time debiasing framework that enforces fairness by filtering generation outputs in real time. |
| Outcome: | The proposed framework mitigates social bias across a range of LLMs while preserving overall generation quality. |
The Impossibility of Fair LLMs (2025.acl-long)
Copied to clipboard
| Challenge: | Existing frameworks for evaluating large language models do not extend to general-purpose AI contexts or are infeasible in practice. |
| Approach: | They analyze a variety of technical fairness frameworks to find inherent challenges . they find that each framework does not logically extend to the general-purpose AI context . |
| Outcome: | The proposed frameworks do not logically extend to the general-purpose AI context or are infeasible in practice due to large amounts of unstructured training data and potential combinations of human populations, use cases, and sensitive attributes. |
FLEX: A Benchmark for Evaluating Robustness of Fairness in Large Language Models (2025.findings-naacl)
Copied to clipboard
| Challenge: | Existing safety evaluations may overlook the inherent weaknesses of Large Language Models, despite their benefits. |
| Approach: | They propose a benchmark to evaluate the robustness of Large Language Models under extreme conditions. |
| Outcome: | The proposed approach evaluates the fairness of large language models under extreme conditions. |
Watching the AI Watchdogs: A Fairness and Robustness Analysis of AI Safety Moderation Classifiers (2025.naacl-short)
Copied to clipboard
| Challenge: | ASM classifiers are designed to moderate content on social media platforms and serve as guardrails that prevent Large Language Models (LLMs) from being fine-tuned on unsafe inputs. |
| Approach: | They examine the fairness and robustness of four widely-used, closed-source ASM classifiers: OpenAI Moderation API, Perspective API, Google Cloud Natural Language (GCNL) API, and Clarifai API. |
| Outcome: | The classifiers do not unfairly classify content belonging to minority groups as unsafe compared to those belonging to majority groups and their behavior remains robust and consistent across similar inputs. |
Preserving Fairness and Safety in Quantized LLMs Through Critical Weight Protection (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing quantization methods focus on general metrics like perplexity or accuracy on standard benchmarks. |
| Approach: | They propose a method that preserves fairness- and safety-critical weights during quantization. |
| Outcome: | The proposed method reduces bias and safety degradation without costly retraining or alignment while maintaining trustworthiness while retaining efficiency. |
Balancing out Bias: Achieving Fairness Through Balanced Training (2022.emnlp-main)
Copied to clipboard
| Challenge: | Existing approaches to reducing group bias do not account for correlations between author demographics and linguistic variables, limiting their effectiveness. |
| Approach: | They extend a method for countering group bias using balanced training by balancing each demographic group in training and using protected attributes as input. |
| Outcome: | The proposed model outperforms all other methods when combined with balanced training. |
Confronting LLMs with Traditional ML: Rethinking the Fairness of Large Language Models in Tabular Classifications (2024.naacl-long)
Copied to clipboard
| Challenge: | Recent studies suggest using large language models to make tabular classifications . however, LLMs have been shown to exhibit harmful social biases based on stereotypes and inequalities present in society. |
| Approach: | They propose to use large language models to make tabular classifications . they show that LLMs inherit biases from their training data . |
| Outcome: | The proposed models exhibit harmful biases that reflect stereotypes and inequalities in society. |
Fair-CCD: Mitigating Bias in Large Language Models for Tabular Classification Through Context-Contrastive Decoding (2026.acl-long)
Copied to clipboard
| Challenge: | Prior work to mitigate fairness issues often employs subjective demonstration selection, leading to low controllability and limited stability across different models and tasks. |
| Approach: | They propose to use in-context learning to insert social biases into large language models to create a structured and controllable representation of the relationship between sensitive attributes and predicted labels. |
| Outcome: | Extensive experiments show that Fair-CCD consistently improves fairness metrics without degrading task accuracy. |