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

Large language models (LLMs) have demonstrated remarkable capabilities invarious tasks. However, their suitability for domain-specific tasks, is limiteddue to their immense scale at deployment, susceptibility to misinformation, andmore importantly, high data annotation costs. We propose a novel InteractiveMulti-Fidelity Learning (IMFL) framework for the cost-effective development ofsmall domain-specific LMs under limited annotation budgets. Our approachformulates the domain-specific fine-tuning process as a multi-fidelity learningproblem, focusing on identifying the optimal acquisition strategy that balancesbetween low-fidelity automatic LLM annotations and high-fidelity humanannotations to maximize model performance. We further propose anexploration-exploitation query strategy that enhances annotation diversity andinformativeness, incorporating two innovative designs: 1) prompt retrieval thatselects in-context examples from human-annotated samples to improve LLMannotation, and 2) variable batch size that controls the order for choosingeach fidelity to facilitate knowledge distillation, ultimately enhancingannotation quality. Extensive experiments on financial and medical tasksdemonstrate that IMFL achieves superior performance compared with singlefidelity annotations. Given a limited budget of human annotation, IMFLsignificantly outperforms the human annotation baselines in all four tasks andachieves very close performance as human annotations on two of the tasks. Thesepromising results suggest that the high human annotation costs indomain-specific tasks can be significantly reduced by employing IMFL, whichutilizes fewer human annotations, supplemented with cheaper and faster LLM(e.g., GPT-3.5) annotations to achieve comparable performance.

Interactive Multi-fidelity Learning for Cost-effective Adaptation of Language Model with Sparse Human Supervision