Saved in:
Bibliographic Details
Main Authors: Guo, Chenqi, Zhong, Shiwei, Liu, Xiaofeng, Feng, Qianli, Ma, Yinglong
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2405.00739
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909187670278144
author Guo, Chenqi
Zhong, Shiwei
Liu, Xiaofeng
Feng, Qianli
Ma, Yinglong
author_facet Guo, Chenqi
Zhong, Shiwei
Liu, Xiaofeng
Feng, Qianli
Ma, Yinglong
contents Does Knowledge Distillation (KD) really work? Conventional wisdom viewed it as a knowledge transfer procedure where a perfect mimicry of the student to its teacher is desired. However, paradoxical studies indicate that closely replicating the teacher's behavior does not consistently improve student generalization, posing questions on its possible causes. Confronted with this gap, we hypothesize that diverse attentions in teachers contribute to better student generalization at the expense of reduced fidelity in ensemble KD setups. By increasing data augmentation strengths, our key findings reveal a decrease in the Intersection over Union (IoU) of attentions between teacher models, leading to reduced student overfitting and decreased fidelity. We propose this low-fidelity phenomenon as an underlying characteristic rather than a pathology when training KD. This suggests that stronger data augmentation fosters a broader perspective provided by the divergent teacher ensemble and lower student-teacher mutual information, benefiting generalization performance. These insights clarify the mechanism on low-fidelity phenomenon in KD. Thus, we offer new perspectives on optimizing student model performance, by emphasizing increased diversity in teacher attentions and reduced mimicry behavior between teachers and student.
format Preprint
id arxiv_https___arxiv_org_abs_2405_00739
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Why does Knowledge Distillation Work? Rethink its Attention and Fidelity Mechanism
Guo, Chenqi
Zhong, Shiwei
Liu, Xiaofeng
Feng, Qianli
Ma, Yinglong
Machine Learning
Computer Vision and Pattern Recognition
Image and Video Processing
Does Knowledge Distillation (KD) really work? Conventional wisdom viewed it as a knowledge transfer procedure where a perfect mimicry of the student to its teacher is desired. However, paradoxical studies indicate that closely replicating the teacher's behavior does not consistently improve student generalization, posing questions on its possible causes. Confronted with this gap, we hypothesize that diverse attentions in teachers contribute to better student generalization at the expense of reduced fidelity in ensemble KD setups. By increasing data augmentation strengths, our key findings reveal a decrease in the Intersection over Union (IoU) of attentions between teacher models, leading to reduced student overfitting and decreased fidelity. We propose this low-fidelity phenomenon as an underlying characteristic rather than a pathology when training KD. This suggests that stronger data augmentation fosters a broader perspective provided by the divergent teacher ensemble and lower student-teacher mutual information, benefiting generalization performance. These insights clarify the mechanism on low-fidelity phenomenon in KD. Thus, we offer new perspectives on optimizing student model performance, by emphasizing increased diversity in teacher attentions and reduced mimicry behavior between teachers and student.
title Why does Knowledge Distillation Work? Rethink its Attention and Fidelity Mechanism
topic Machine Learning
Computer Vision and Pattern Recognition
Image and Video Processing
url https://arxiv.org/abs/2405.00739