Feature Structure Distillation with Centered Kernel Alignment in BERT Transferring

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jung, Hee-Jun, Kim, Doyeon, Na, Seung-Hoon, Kim, Kangil
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910668309921792
author Jung, Hee-Jun
Kim, Doyeon
Na, Seung-Hoon
Kim, Kangil
author_facet Jung, Hee-Jun
Kim, Doyeon
Na, Seung-Hoon
Kim, Kangil
contents Knowledge distillation is an approach to transfer information on representations from a teacher to a student by reducing their difference. A challenge of this approach is to reduce the flexibility of the student's representations inducing inaccurate learning of the teacher's knowledge. To resolve it in transferring, we investigate distillation of structures of representations specified to three types: intra-feature, local inter-feature, global inter-feature structures. To transfer them, we introduce feature structure distillation methods based on the Centered Kernel Alignment, which assigns a consistent value to similar features structures and reveals more informative relations. In particular, a memory-augmented transfer method with clustering is implemented for the global structures. The methods are empirically analyzed on the nine tasks for language understanding of the GLUE dataset with Bidirectional Encoder Representations from Transformers (BERT), which is a representative neural language model. In the results, the proposed methods effectively transfer the three types of structures and improve performance compared to state-of-the-art distillation methods. Indeed, the code for the methods is available in https://github.com/maroo-sky/FSD.
format Preprint
id arxiv_https___arxiv_org_abs_2204_08922
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Feature Structure Distillation with Centered Kernel Alignment in BERT Transferring
Jung, Hee-Jun
Kim, Doyeon
Na, Seung-Hoon
Kim, Kangil
Computation and Language
Artificial Intelligence
Machine Learning
Knowledge distillation is an approach to transfer information on representations from a teacher to a student by reducing their difference. A challenge of this approach is to reduce the flexibility of the student's representations inducing inaccurate learning of the teacher's knowledge. To resolve it in transferring, we investigate distillation of structures of representations specified to three types: intra-feature, local inter-feature, global inter-feature structures. To transfer them, we introduce feature structure distillation methods based on the Centered Kernel Alignment, which assigns a consistent value to similar features structures and reveals more informative relations. In particular, a memory-augmented transfer method with clustering is implemented for the global structures. The methods are empirically analyzed on the nine tasks for language understanding of the GLUE dataset with Bidirectional Encoder Representations from Transformers (BERT), which is a representative neural language model. In the results, the proposed methods effectively transfer the three types of structures and improve performance compared to state-of-the-art distillation methods. Indeed, the code for the methods is available in https://github.com/maroo-sky/FSD.
title Feature Structure Distillation with Centered Kernel Alignment in BERT Transferring
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2204.08922