Uncertainty Estimation in Autoregressive Structured Prediction

By Andrey Malinin, Mark John Francis Gales. International Conference on Learning Representations, 2021

509 citations recorded by the source corpus.

View Paper: https://www.semanticscholar.org/paper/0921322cf6ea34d1852f13cb67eeac9d1f863518

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

Citations in the corpus

Detecting hallucinations in large language models using semantic entropy (2024, unknown): “In expectation, longer sequences have lower joint likelihoods because of the conditional independence of the token probabilities 47 .”
INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection (2024, methodology): “Compared to existing uncertainty or consistency metrics that obtained in logit or language space (Malinin & Gales, 2020; Huang et al., 2023; Lin et al., 2022), the advantages of EigenScore are: (1) It captures the semantic divergence (entropy) in the dense embedding space, which is expected to…”
Generating with Confidence: Uncertainty Quantification for Black-box Large Language Models (2023, methodology): “As existing uncertainty quantification research for NLG primarily focuses on input uncertainty (Kuhn et al., 2023; Malinin & Gales, 2021), it overlooks the crucial aspect of confidence, which is essential in deciding when to trust an LLM’s response (see Section 3.2 for more discussion).”
SelfCheckGPT: Zero-Resource Black-Box Hallucination Detection for Generative Large Language Models (2023, unknown): “For example, OpenAI’s GPT-3 web interface allows users to display token probabilities (as shown in Figure 2), and further uncertainty estimation approaches based on aleatoric and epistemic uncertainty have been studied for autoregressive generation (Xiao and Wang, 2021; Malinin and Gales, 2021).”
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions (2023, unknown): “Moreover, some works (Malinin and Gales, 2021) have sought to quantify model uncertainty using log probability.”
RAGTruth: A Hallucination Corpus for Developing Trustworthy Retrieval-Augmented Language Models (2023, background): “In contentious cases, additional annotations are provided to accu-312 rately reflect these situations.”
CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing (2023, methodology): “Norm Entropy (Malinin & Gales, 2021) leverages geometric mean token probability, where we calculate confidence as the arithmetic mean negative log-probability, given by ConfNorm Entropy = − 1N ∑N i log p(yi|y i).”
Can LLMs Express Their Uncertainty? An Empirical Evaluation of Confidence Elicitation in LLMs (2023, methodology): “…confidence from machine learning models has predominantly relied on white-box access to internal model information, such as token-likelihoods (Malinin & Gales, 2020; Kadavath et al., 2022) and associated calibration techniques (Jiang et al., 2021), as well as model fine-tuning (Lin et al.,…”
Semantic Uncertainty: Linguistic Invariances for Uncertainty Estimation in Natural Language Generation (2023, methodology): “Normalised entropy reimplements single-model variant of Malinin & Gales (2020), lexical similarity measures the average Rouge-L in a sampled set of answers for a given question analogously to Fomicheva et al. (2020), entropy and p ( True ) reimplement Kadavath et al. (2022).”
Embedding Trust: Semantic Isotropy Predicts Nonfactuality in Long-Form Text Generation (2025, unknown): “On the other hand, Monte Carlo sampling-based approaches (Zhang, 2020; Malinin & Gales, 2021; Lin et al., 2024; Lakshminarayanan et al., 2017) have shown promise in estimating uncertainty in LLMs.”
LLM-confidence reranker: A training-free approach for enhancing retrieval-augmented generation systems (2026, unknown): “To enable comparison across different length outputs, Malinin & Gales (2021) systematically applied length-normalization to various information-theoretic uncertainty measures, enhancing their practical utility.”