The Privileged Students: On the Value of Initialization in Multilingual Knowledge Distillation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wibowo, Haryo Akbarianto, Solorio, Thamar, Aji, Alham Fikri
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913574035652608
author Wibowo, Haryo Akbarianto
Solorio, Thamar
Aji, Alham Fikri
author_facet Wibowo, Haryo Akbarianto
Solorio, Thamar
Aji, Alham Fikri
contents Knowledge distillation (KD) has proven to be a successful strategy to improve the performance of smaller models in many NLP tasks. However, most of the work in KD only explores monolingual scenarios. In this paper, we investigate the value of KD in multilingual settings. We find the significance of KD and model initialization by analyzing how well the student model acquires multilingual knowledge from the teacher model. Our proposed method emphasizes copying the teacher model's weights directly to the student model to enhance initialization. Our findings show that model initialization using copy-weight from the fine-tuned teacher contributes the most compared to the distillation process itself across various multilingual settings. Furthermore, we demonstrate that efficient weight initialization preserves multilingual capabilities even in low-resource scenarios.
format Preprint
id arxiv_https___arxiv_org_abs_2406_16524
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle The Privileged Students: On the Value of Initialization in Multilingual Knowledge Distillation
Wibowo, Haryo Akbarianto
Solorio, Thamar
Aji, Alham Fikri
Computation and Language
68T50
Knowledge distillation (KD) has proven to be a successful strategy to improve the performance of smaller models in many NLP tasks. However, most of the work in KD only explores monolingual scenarios. In this paper, we investigate the value of KD in multilingual settings. We find the significance of KD and model initialization by analyzing how well the student model acquires multilingual knowledge from the teacher model. Our proposed method emphasizes copying the teacher model's weights directly to the student model to enhance initialization. Our findings show that model initialization using copy-weight from the fine-tuned teacher contributes the most compared to the distillation process itself across various multilingual settings. Furthermore, we demonstrate that efficient weight initialization preserves multilingual capabilities even in low-resource scenarios.
title The Privileged Students: On the Value of Initialization in Multilingual Knowledge Distillation
topic Computation and Language
68T50
url https://arxiv.org/abs/2406.16524