English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Uncertainty decomposition refers to the task of decomposing the totaluncertainty of a model into data (aleatoric) uncertainty, resulting from theinherent complexity or ambiguity of the data, and model (epistemic)uncertainty, resulting from the lack of knowledge in the model. Performinguncertainty decomposition for large language models (LLMs) is an important steptoward improving the reliability, trustworthiness, and interpretability ofLLMs, but this research task is very challenging and remains unresolved. Theexisting canonical method, Bayesian Neural Network (BNN), cannot be applied toLLMs, because BNN requires training and ensembling multiple variants of models,which is infeasible or prohibitively expensive for LLMs. In this paper, weintroduce an uncertainty decomposition framework for LLMs, called inputclarifications ensemble, which bypasses the need to train new models. Ratherthan ensembling models with different parameters, our approach generates a setof clarifications for the input, feeds them into the fixed LLMs, and ensemblesthe corresponding predictions. We show that our framework shares a symmetricdecomposition structure with BNN. Empirical evaluations demonstrate that theproposed framework provides accurate and reliable uncertainty quantification onvarious tasks. Code will be made publicly available athttps://github.com/UCSB-NLP-Chang/llm_uncertainty .

Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling