Enregistré dans:
Détails bibliographiques
Auteurs principaux: Lo, Ka Man, Liang, Yiming, Du, Wenyu, Fan, Yuantao, Wang, Zili, Huang, Wenhao, Ma, Lei, Fu, Jie
Format: Preprint
Publié: 2024
Sujets:
Accès en ligne:https://arxiv.org/abs/2402.16918
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914860365774848
author Lo, Ka Man
Liang, Yiming
Du, Wenyu
Fan, Yuantao
Wang, Zili
Huang, Wenhao
Ma, Lei
Fu, Jie
author_facet Lo, Ka Man
Liang, Yiming
Du, Wenyu
Fan, Yuantao
Wang, Zili
Huang, Wenhao
Ma, Lei
Fu, Jie
contents Modular neural architectures are gaining attention for their powerful generalization and efficient adaptation to new domains. However, training these models poses challenges due to optimization difficulties arising from intrinsic sparse connectivity. Leveraging knowledge from monolithic models through techniques like knowledge distillation can facilitate training and enable integration of diverse knowledge. Nevertheless, conventional knowledge distillation approaches are not tailored to modular models and struggle with unique architectures and enormous parameter counts. Motivated by these challenges, we propose module-to-module knowledge distillation (m2mKD) for transferring knowledge between modules. m2mKD combines teacher modules of a pretrained monolithic model and student modules of a modular model with a shared meta model respectively to encourage the student module to mimic the behaviour of the teacher module. We evaluate m2mKD on two modular neural architectures: Neural Attentive Circuits (NACs) and Vision Mixture-of-Experts (V-MoE). Applying m2mKD to NACs yields significant improvements in IID accuracy on Tiny-ImageNet (up to 5.6%) and OOD robustness on Tiny-ImageNet-R (up to 4.2%). Additionally, the V-MoE-Base model trained with m2mKD achieves 3.5% higher accuracy than end-to-end training on ImageNet-1k. Code is available at https://github.com/kamanphoebe/m2mKD.
format Preprint
id arxiv_https___arxiv_org_abs_2402_16918
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle m2mKD: Module-to-Module Knowledge Distillation for Modular Transformers
Lo, Ka Man
Liang, Yiming
Du, Wenyu
Fan, Yuantao
Wang, Zili
Huang, Wenhao
Ma, Lei
Fu, Jie
Machine Learning
Computer Vision and Pattern Recognition
Modular neural architectures are gaining attention for their powerful generalization and efficient adaptation to new domains. However, training these models poses challenges due to optimization difficulties arising from intrinsic sparse connectivity. Leveraging knowledge from monolithic models through techniques like knowledge distillation can facilitate training and enable integration of diverse knowledge. Nevertheless, conventional knowledge distillation approaches are not tailored to modular models and struggle with unique architectures and enormous parameter counts. Motivated by these challenges, we propose module-to-module knowledge distillation (m2mKD) for transferring knowledge between modules. m2mKD combines teacher modules of a pretrained monolithic model and student modules of a modular model with a shared meta model respectively to encourage the student module to mimic the behaviour of the teacher module. We evaluate m2mKD on two modular neural architectures: Neural Attentive Circuits (NACs) and Vision Mixture-of-Experts (V-MoE). Applying m2mKD to NACs yields significant improvements in IID accuracy on Tiny-ImageNet (up to 5.6%) and OOD robustness on Tiny-ImageNet-R (up to 4.2%). Additionally, the V-MoE-Base model trained with m2mKD achieves 3.5% higher accuracy than end-to-end training on ImageNet-1k. Code is available at https://github.com/kamanphoebe/m2mKD.
title m2mKD: Module-to-Module Knowledge Distillation for Modular Transformers
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2402.16918