DC Field | Value | Language |
---|---|---|
dc.contributor.author | Myung, Sanghoon | ko |
dc.contributor.author | Huh, In | ko |
dc.contributor.author | Jang, Wonik | ko |
dc.contributor.author | Choe, Jae Myung | ko |
dc.contributor.author | Ryu, Jisu | ko |
dc.contributor.author | Kim, Dae Sin | ko |
dc.contributor.author | Kim, Kee-Eung | ko |
dc.contributor.author | Jeong, Changwook | ko |
dc.date.accessioned | 2023-09-21T02:00:36Z | - |
dc.date.available | 2023-09-21T02:00:36Z | - |
dc.date.created | 2023-09-21 | - |
dc.date.issued | 2022-07 | - |
dc.identifier.citation | 39th International Conference on Machine Learning, ICML 2022, pp.16240 - 16252 | - |
dc.identifier.issn | 2640-3498 | - |
dc.identifier.uri | http://hdl.handle.net/10203/312793 | - |
dc.description.abstract | Inductive transfer learning aims to learn from a small amount of training data for the target task by utilizing a pre-trained model from the source task. Most strategies that involve large-scale deep learning models adopt initialization with the pretrained model and fine-tuning for the target task. However, when using over-parameterized models, we can often prune the model without sacrificing the accuracy of the source task. This motivates us to adopt model pruning for transfer learning with deep learning models. In this paper, we propose PAC-Net, a simple yet effective approach for transfer learning based on pruning. PAC-Net consists of three steps: Prune, Allocate, and Calibrate (PAC). The main idea behind these steps is to identify essential weights for the source task, fine-tune on the source task by updating the essential weights, and then calibrate on the target task by updating the remaining redundant weights. Under the various and extensive set of inductive transfer learning experiments, we show that our method achieves state-of-the-art performance by a large margin. | - |
dc.language | English | - |
dc.publisher | ML Research Press | - |
dc.title | PAC-Net: A Model Pruning Approach to Inductive Transfer Learning | - |
dc.type | Conference | - |
dc.identifier.wosid | 000900064906015 | - |
dc.identifier.scopusid | 2-s2.0-85144269799 | - |
dc.type.rims | CONF | - |
dc.citation.beginningpage | 16240 | - |
dc.citation.endingpage | 16252 | - |
dc.citation.publicationname | 39th International Conference on Machine Learning, ICML 2022 | - |
dc.identifier.conferencecountry | US | - |
dc.identifier.conferencelocation | Baltimore, MD | - |
dc.contributor.localauthor | Kim, Kee-Eung | - |
dc.contributor.nonIdAuthor | Myung, Sanghoon | - |
dc.contributor.nonIdAuthor | Huh, In | - |
dc.contributor.nonIdAuthor | Jang, Wonik | - |
dc.contributor.nonIdAuthor | Choe, Jae Myung | - |
dc.contributor.nonIdAuthor | Ryu, Jisu | - |
dc.contributor.nonIdAuthor | Kim, Dae Sin | - |
dc.contributor.nonIdAuthor | Jeong, Changwook | - |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.