Model Uncertainty-Aware Knowledge Amalgamation for Pre-Trained Language Models

TL;DR

This paper introduces MUKA, a novel framework for merging knowledge from multiple pre-trained language models without human annotations, improving model reuse efficiency and generalization across various settings.

Contribution

The paper proposes MUKA, a model uncertainty-aware approach for knowledge amalgamation from multiple teacher PLMs without annotations, enhancing reuse and generalization.

Findings

MUKA outperforms baseline methods on benchmark datasets.

MUKA generalizes well with multiple, heterogeneous, and cross-dataset teachers.

Experimental results confirm the effectiveness of the proposed framework.

Abstract

As many fine-tuned pre-trained language models~(PLMs) with promising performance are generously released, investigating better ways to reuse these models is vital as it can greatly reduce the retraining computational cost and the potential environmental side-effects. In this paper, we explore a novel model reuse paradigm, Knowledge Amalgamation~(KA) for PLMs. Without human annotations available, KA aims to merge the knowledge from different teacher-PLMs, each of which specializes in a different classification problem, into a versatile student model. The achieve this, we design a Model Uncertainty--aware Knowledge Amalgamation~(MUKA) framework, which identifies the potential adequate teacher using Monte-Carlo Dropout for approximating the golden supervision to guide the student. Experimental results demonstrate that MUKA achieves substantial improvements over baselines on benchmark datasets. Further analysis shows that MUKA can generalize well under several complicate settings with multiple teacher models, heterogeneous teachers, and even cross-dataset teachers.