The Effects of Data Augmentation on Confidence Estimation for LLMs

2025/05/21 by Rui Wang, Wang, Rui, Zhu, Renyu +9
Computer Science · Decision Sciences · #Advanced Data Storage Technologies #Distributed and Parallel Computing Systems #FOS: Computer and information sciences #Machine Learning (cs.LG) #Simulation Techniques and Applications

paper · pdf · doi:10.48550/arxiv.2506.11046

openalex publication_date 2025/05/21 · openalex created_date 2025/10/11 · openalex updated_date 2026/07/28

Abstract

Confidence estimation is crucial for reflecting the reliability of large language models (LLMs), particularly in the widely used closed-source models. Utilizing data augmentation for confidence estimation is viable, but discussions focus on specific augmentation techniques, limiting its potential. We study the impact of different data augmentation methods on confidence estimation. Our findings indicate that data augmentation strategies can achieve better performance and mitigate the impact of overconfidence. We investigate the influential factors related to this and discover that, while preserving semantic information, greater data diversity enhances the effectiveness of augmentation. Furthermore, the impact of different augmentation strategies varies across different range of application. Considering parameter transferability and usability, the random combination of augmentations is a promising choice.

Citations

• Calibrating Language Models with Adaptive Temperature Scaling
• Gemma 2: Improving Open Language Models at a Practical Size
• Are Data Augmentation Methods in Named Entity Recognition Applicable for Uncertainty Estimation?
• Enhancing Language Model Factuality via Activation-Based Confidence Calibration and Guided Decoding
• Confidence Under the Hood: An Investigation into the Confidence-Probability Alignment in Large Language Models
• SAC3: Reliable Hallucination Detection in Black-Box Language Models via Semantic-aware Cross-check Consistency
• Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting
• Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs
• Just Ask for Calibration: Strategies for Eliciting Calibrated Confidence Scores from Language Models Fine-Tuned with Human Feedback
• Prompting GPT-3 To Be Reliable
• Teaching Models to Express Their Uncertainty in Words
• Large Language Models are Zero-Shot Reasoners
• PaLM: Scaling Language Modeling with Pathways
• Training Verifiers to Solve Math Word Problems
• Text AutoAugment: Learning Compositional Augmentation Policy for Text Classification
• A Survey of Data Augmentation Approaches for NLP
• Did Aristotle Use a Laptop? A Question Answering Benchmark with Implicit Reasoning Strategies
• Measuring Massive Multitask Language Understanding
• RandAugment: Practical automated data augmentation with a reduced search space
• EDA: Easy Data Augmentation Techniques for Boosting Performance on Text Classification Tasks
• Improving Neural Machine Translation Models with Monolingual Data

Related

• A Survey of Confidence Estimation and Calibration in Large Language Models
• Large Language Model Confidence Estimation via Black-Box Access
• SteerConf: Steering LLMs for Confidence Elicitation
• Mind the Generation Process: Fine-Grained Confidence Estimation During LLM Generation
• Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs