2026-09-20T22:16:12Z https://www.ideals.illinois.edu/oai-pmh

oai:www.ideals.illinois.edu:2142/129259 2025-10-20 col_2142_5131 col_2142_10761 com_2142_5130 com_2142_10755 com_2142_234

application/pdf en text Submission original under an indefinite embargo labeled 'Open Access'. The submission was exported from vireo on 2025-10-19 without embargo terms The student, Rishabh Adiga, accepted the attached license on 2025-04-26 at 09:28. The student, Rishabh Adiga, submitted this Thesis for approval on 2025-04-26 at 09:34. This Thesis was approved for publication on 2025-04-28 at 18:25. DSpace SAF Submission Ingestion Package generated from Vireo submission #22007 on 2025-10-19 at 18:11:05 Enhancing large language models: toward more reliable and equitable NLP Adiga, Rishabh 2025-04-28 Chandrasekaran, Varun LLM Few-shot Attention Bias Prompting Fairness Metrics Localization Mitigation eng Recent advances in large language models (LLMs) have enabled them to achieve striking fewshot performance on complex tasks, yet they can remain sensitive to prompt configurations and prone to subtle biases. This thesis addresses two central problems: (1) selecting informative few-shot examples and (2) localizing and mitigating bias in ambiguous comparative prompts. First, we propose a complexity-based approach for selecting examples in few-shot sequence tagging tasks, aiming to align test examples with training examples based on syntactic and semantic metrics. By focusing on features like sentence similarity, length matching, and label diversity, we achieve more consistent and robust outcomes without fine-tuning or adding parameters. Second, we develop a method to localize and mitigate bias in LLMs by examining the attention layers that favor certain entities over others. We then scale attention in those identified layers to reduce skewed preferences, preserving overall model fluency while mitigating biases. Extensive evaluations confirm that this targeted attention manipulation provides a lightweight way to address fairness concerns without sacrificing downstream accuracy. 2025-05 Thesis https://hdl.handle.net/2142/129259 © 2025 Rishabh Adiga Siebel School Comp & Data Sci Computer Science University of Illinois Urbana-Champaign M.S. Thesis