Context-aware Fairness Evaluation and Mitigation in LLMs
This addresses fairness and safety issues in conversational AI, offering an incremental improvement over static pruning methods by enabling dynamic adaptation to changing contexts.
The paper tackles the problem of undesirable behaviors in large language models, such as unfairness and harmful content amplification, by proposing a dynamic, reversible pruning-based framework that adaptively masks neurons during generation, achieving fine-grained mitigation and more coherent behavior across multilingual dialogues.
Large language models often display undesirable behaviors embedded in their internal representations, undermining fairness, inconsistency drift, amplification of harmful content, and the propagation of unwanted patterns during extended dialogue and conversations. Although training-time or data-centric methods attempt to reduce these effects, they are computationally expensive, irreversible once deployed, and slow to adapt to new conversational contexts. Pruning-based methods provide a flexible and transparent way to reduce bias by adjusting the neurons responsible for certain behaviors. However, most existing approaches are static; once a neuron is removed, the model loses the ability to adapt when the conversation or context changes. To address this, we propose a dynamic, reversible, pruning-based framework that detects context-aware neuron activations and applies adaptive masking to modulate their influence during generation. Our inference-time solution provides fine-grained, memory-aware mitigation with knowledge-preserved, more coherent behavior across multilingual single- and multi-turn dialogues, enabling dynamic fairness control in real-world conversational AI.