Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
February 21, 2026IEEE Transactions on Neural Networks and Learning Systems

Dual-Teacher and Dual-Prompt Pool for Few-Shot Dialog State Tracking

View Full Paper
Ask AI
Bookmark
Share

Authors

DWDi WuYLyuheng liZYZhizhi Yu

Discussion

Loading...

Member takes

Overview

This work develops a dual-teacher and dual-prompt model for few-shot dialog state tracking, enhancing performance in dialogue systems.

Key Points

  • The aim is to improve few-shot dialog state tracking (DST) by leveraging semantic information and enhancing adaptability.
  • Developed dual-teacher models for generating pseudolabels.
  • Employed self-training to enhance state value generation.
  • Designed a dual-prompt fine-tuning strategy.
  • Constructed a dynamic prompt pool for adaptive prompt generation.
  • Incorporated reconstruction errors into the model for improved accuracy.
  • DDP-DST outperforms baseline models like SM2-3b, DS2, and SVAG.
  • Achieved average improvements of 4.3%, 2.4%, and 2.0% in joint goal accuracy.
  • Maintained competitive performance with fewer than 1 billion parameters.

Cite This Study

Wu et al. (2026) studied this question.

synapsesocial.com/papers/69994bdd873532290d01fe17https://doi.org/10.1109/tnnls.2026.3659341
View Full Paper
Ask AI
Bookmark
Share