This page organizes papers related to artificial intelligence published around the world. This page is summarized using Google Gemini and is operated on a non-profit basis. The copyright of the paper belongs to the author and the relevant institution. When sharing, simply cite the source.
HiCoLoRA: Addressing Context-Prompt Misalignment via Hierarchical Collaborative LoRA for Zero-Shot DST
Created by
Haebom
Author
Shuyu Zhang, Yifan Wei, Xinru Wang, Yanmin Zhu, Yangfan He, Yixuan Weng, Bin Li
Outline
This paper proposes a Hierarchical Collaborative Low-Rank Adaptation (HiCoLoRA) framework that enhances zero-shot slot inference through robust prompt alignment to address the challenge of Zero-shot Dialog State Tracking (zs-DST) for Task-Oriented Dialog Systems (TODs) that generalizes to new domains without data annotation. HiCoLoRA features a hierarchical LoRA architecture for dynamic layer-by-layer processing, Spectral Joint Domain-Slot Clustering to identify transferable associations, and Semantic-Enhanced SVD Initialization (SemSVD-Init) to preserve pre-trained knowledge. Experimental results on the MultiWOZ and SGD datasets demonstrate that HiCoLoRA achieves state-of-the-art performance on zs-DST and outperforms existing baselines.
Takeaways, Limitations
•
Takeaways:
◦
Proposing a new framework (HiCoLoRA) to solve the zero-shot DST problem.
◦
Introducing innovative methodologies such as hierarchical LoRA architecture, Spectral Joint Domain-Slot Clustering, and SemSVD-Init.
◦
Achieving SOTA on MultiWOZ and SGD datasets
◦
Code disclosure
•
Limitations:
◦
The specific Limitations of the paper is not provided (only the Abstract content is included)
