多任务学习在分子性质预测中的对比研究
随着深度学习的快速发展,相关算法被广泛应用于量子化学计算领域以实现高效的分子设计及性质研究.其中,多任务学习方法通过挖掘分子性质之间的关系可以同时预测多个分子属性,然而此类研究目前较为有限.本文采用硬参数共享结构与损失函数加权方法来实现多任务分子性质预测.通过对比单任务基准与各类多任务模型在不同分子属性集上的性能,展示了多属性预测精度强烈依赖于属性间的关系,当关联变复杂时,硬参数共享可以提高预测精度.此外,恰当的损失函数加权方法有利于实现更均衡的多目标优化,使预测更准确.进一步的实验展示了多任务学习模型的计算效率优势及其在训练数据量受限时的预测性能优势....
Saved in:
| Published in | 化学物理学报（英文版） Vol. 36; no. 4; pp. 443 - 452 |
|---|---|
| Main Authors | , , , , |
| Format | Journal Article |
| Language | Chinese |
| Published |
中国科学技术大学物理学院,合肥 230026%中国科学技术大学计算机科学与技术学院,大数据分析与应用安徽省重点实验室,合肥 230026
2023
|
| Subjects | |
| Online Access | Get full text |
| ISSN | 1674-0068 |
| DOI | 10.1063/1674-0068/cjcp2203055 |
Cover
| Abstract | 随着深度学习的快速发展,相关算法被广泛应用于量子化学计算领域以实现高效的分子设计及性质研究.其中,多任务学习方法通过挖掘分子性质之间的关系可以同时预测多个分子属性,然而此类研究目前较为有限.本文采用硬参数共享结构与损失函数加权方法来实现多任务分子性质预测.通过对比单任务基准与各类多任务模型在不同分子属性集上的性能,展示了多属性预测精度强烈依赖于属性间的关系,当关联变复杂时,硬参数共享可以提高预测精度.此外,恰当的损失函数加权方法有利于实现更均衡的多目标优化,使预测更准确.进一步的实验展示了多任务学习模型的计算效率优势及其在训练数据量受限时的预测性能优势. |
|---|---|
| AbstractList | 随着深度学习的快速发展,相关算法被广泛应用于量子化学计算领域以实现高效的分子设计及性质研究.其中,多任务学习方法通过挖掘分子性质之间的关系可以同时预测多个分子属性,然而此类研究目前较为有限.本文采用硬参数共享结构与损失函数加权方法来实现多任务分子性质预测.通过对比单任务基准与各类多任务模型在不同分子属性集上的性能,展示了多属性预测精度强烈依赖于属性间的关系,当关联变复杂时,硬参数共享可以提高预测精度.此外,恰当的损失函数加权方法有利于实现更均衡的多目标优化,使预测更准确.进一步的实验展示了多任务学习模型的计算效率优势及其在训练数据量受限时的预测性能优势. |
| Abstract_FL | With the bloom of deep learning algorithms,various models have been widely utilized in quantum chemistry cal-culation to design new molecules and explore molecular properties.However,limited stud-ies focus on multi-task molecular property prediction,which offers more efficient ways to si-multaneously learn different but related properties by leveraging the inter-task relationship.In this work,we apply the hard parameter sharing framework and advanced loss weighting methods to multi-task molecular property prediction.Based on the performance comparison between single-task baseline and multi-task models on several task sets,we find that the pre-diction accuracy largely depends on the inter-task relationship,and hard parameter sharing improves the performance when the correlation becomes complex.In addition,we show that proper loss weighting methods help achieve more balanced multi-task optimization and en-hance the prediction accuracy.Our additional experiments on varying amount of training da-ta further validate the multi-task advantages and show that multi-task models with proper loss weighting methods can achieve more accurate prediction of molecular properties with much less computational cost. |
| Author | 朱健保 刘淇 王皓 朱文光 韩超 |
| AuthorAffiliation | 中国科学技术大学物理学院,合肥 230026%中国科学技术大学计算机科学与技术学院,大数据分析与应用安徽省重点实验室,合肥 230026 |
| AuthorAffiliation_xml | – name: 中国科学技术大学物理学院,合肥 230026%中国科学技术大学计算机科学与技术学院,大数据分析与应用安徽省重点实验室,合肥 230026 |
| Author_FL | Chao Han Jianbao Zhu Wenguang Zhu Hao Wang Qi Liu |
| Author_FL_xml | – sequence: 1 fullname: Chao Han – sequence: 2 fullname: Hao Wang – sequence: 3 fullname: Jianbao Zhu – sequence: 4 fullname: Qi Liu – sequence: 5 fullname: Wenguang Zhu |
| Author_xml | – sequence: 1 fullname: 韩超 – sequence: 2 fullname: 王皓 – sequence: 3 fullname: 朱健保 – sequence: 4 fullname: 刘淇 – sequence: 5 fullname: 朱文光 |
| BookMark | eNrjYmDJy89LZWCQNTTQMzQwM9Y3NDM30TUwMLPQT85KLjAyMjA2MDVlYeCEi3Mw8BYXZyYZGJkZm5uZWVhwMpg9XTLrye7dT7sWPl277MnOBU_nrHja0fZ07YRnDctfbFnxclHLs63dT3asfT6r5en6nc_WT3m-YMrzldt4GFjTEnOKU3mhNDeDuptriLOHbnliXlpiXnp8Vn5pUR5QJj6jojynIsnIwMjYwMTAwNKYeJUA0bFUGA |
| ContentType | Journal Article |
| Copyright | Copyright © Wanfang Data Co. Ltd. All Rights Reserved. |
| Copyright_xml | – notice: Copyright © Wanfang Data Co. Ltd. All Rights Reserved. |
| DBID | 2B. 4A8 92I 93N PSX TCJ |
| DOI | 10.1063/1674-0068/cjcp2203055 |
| DatabaseName | Wanfang Data Journals - Hong Kong WANFANG Data Centre Wanfang Data Journals 万方数据期刊 - 香港版 China Online Journals (COJ) China Online Journals (COJ) |
| DatabaseTitleList | |
| DeliveryMethod | fulltext_linktorsrc |
| Discipline | Chemistry |
| DocumentTitle_FL | Comparison of Multi-task Approaches on Molecular Property Prediction |
| EndPage | 452 |
| ExternalDocumentID | hxwlxb202304009 |
| GroupedDBID | -SB -S~ 02O 042 1JI 1WK 2B. 4.4 4A8 5B3 5VS 5XA 5XC 7.M 92E 92I 93N AAAAW AABDS AAGWI AAPUP AAYIH ABJGX ABJNI ACAFW ACBRY ACWPO ACZLF ADCTM ADMLS AEJMO AENEX AFHCQ AFYNE AGKCL AGLKD AGTJO AHSDT ALMA_UNASSIGNED_HOLDINGS ATQHT ATXIE AWQPM BBWZM BPZLN CAJEB CW9 DU5 EBS EDWGO EJD EQZZN FDOHQ FFFMQ IOP KOT M45 M71 N9A NT- NT. PSX Q-- RIN RIP RIV RNS RQS SY9 TCJ TGP U1G U5L |
| ID | FETCH-wanfang_journals_hxwlxb2023040093 |
| ISSN | 1674-0068 |
| IngestDate | Thu May 29 04:09:57 EDT 2025 |
| IsPeerReviewed | true |
| IsScholarly | true |
| Issue | 4 |
| Keywords | 分子属性预测 Deep learning 深度学习 多任务学习 Molecular property prediction 损失函数加权方法 Multi-task learning Uncertainty weighting |
| Language | Chinese |
| LinkModel | OpenURL |
| MergedId | FETCHMERGED-wanfang_journals_hxwlxb2023040093 |
| ParticipantIDs | wanfang_journals_hxwlxb202304009 |
| PublicationCentury | 2000 |
| PublicationDate | 2023 |
| PublicationDateYYYYMMDD | 2023-01-01 |
| PublicationDate_xml | – year: 2023 text: 2023 |
| PublicationDecade | 2020 |
| PublicationTitle | 化学物理学报（英文版） |
| PublicationTitle_FL | Chinese Journal of Chemical Physics |
| PublicationYear | 2023 |
| Publisher | 中国科学技术大学物理学院,合肥 230026%中国科学技术大学计算机科学与技术学院,大数据分析与应用安徽省重点实验室,合肥 230026 |
| Publisher_xml | – name: 中国科学技术大学物理学院,合肥 230026%中国科学技术大学计算机科学与技术学院,大数据分析与应用安徽省重点实验室,合肥 230026 |
| SSID | ssib026376688 ssib026376687 ssj0065100 ssib050169557 ssib055692153 ssib004869571 ssib025541138 ssib051367561 ssib023373770 ssib051380371 |
| Score | 4.757357 |
| Snippet | 随着深度学习的快速发展,相关算法被广泛应用于量子化学计算领域以实现高效的分子设计及性质研究.其中,多任务学习方法通过挖掘分子性质之间的关系可以同时预测多个分子属... |
| SourceID | wanfang |
| SourceType | Aggregation Database |
| StartPage | 443 |
| Title | 多任务学习在分子性质预测中的对比研究 |
| URI | https://d.wanfangdata.com.cn/periodical/hxwlxb202304009 |
| Volume | 36 |
| hasFullText | 1 |
| inHoldings | 1 |
| isFullTextHit | |
| isPrint | |
| journalDatabaseRights | – providerCode: PRVEBS databaseName: Inspec with Full Text issn: 1674-0068 databaseCode: ADMLS dateStart: 20080227 customDbUrl: isFulltext: true dateEnd: 20240930 titleUrlDefault: https://www.ebsco.com/products/research-databases/inspec-full-text omitProxy: false ssIdentifier: ssj0065100 providerName: EBSCOhost |
| link | http://utb.summon.serialssolutions.com/2.0.0/link/0/eLvHCXMwrV3NaxQxFA91e9CL-Inf9GDwIGvnK5nkOLM7pYj1YoXeSmc7axFZRVssPXkoInrx0FPxUEW0HkovKpb-NUKn1f_C38tkZ4a62CrCMmTyXt43eS9hkmXsaqY8V2azbnM26zrNwHVlU_vZbNOXCtURSrqu2RqYuC3H7wY3p8TUUON77aulhfn0Rmdp4LmSf_Eq-uBXOiX7F54tiaIDbfgXT3gYz0P5mCeCRwHXEU8CHsf0Q4-KeOQaUJtH0oA0jxzq0S0eKYOjuJIWRwMk6YuHKOSJ4nFgcDDE4yogUAz82NBRhJ-ExJFAGD5GxIGDhg4IBEa2oXks67Wv4dviWtZkC7nShEk0nUokAkmjiOAJWLRIYMgGMWKXQCCiQjucQAVOualB4usxQxgKAVdUEIxJjD5GDe1XEEn2iY3plGtYQ2Xo1a5QjOW0MlYBobC-aVKcaDYBXjMWjA6hjdVgX-3u11A5ljEsWLiT3HAYA0FF4ySvZbg4fRt5EP061nwO_a-m-H-igEhiIiukhg77BosGETRMYejBlAcpYNkBTfC4CMk2MaqiFaDERGtJXBD3Itx0YEMbQ2AyAiVWWfCFP8EOIaNMD2wUa4sMmhTsEcVm0QNJ_mDTWhqXYdCk00_1PO_L2nwW1JJ2UFwUZuu_oLhR-bfSArU87XL1KaPdud955Hnmkr2qmiq_cZ1bfPpgMaXgozypj7BhL0Qx3GDDUXvi1p0qaSqpRbUm83w_9MMqiWLJjxm0SpKeREkgqzWVfS_hgq5XEtWnDILuZKwdXse7cmp3YAohNcr-snyVSNLmFHlf0f6xR-mPln2jNdXNacded6Z3r1aYT55gx-2KeiQqpseTbGhp7hQ72ur_keVpJvP3qzvb2_nLt_nGh52ttfzNev7ieb7xevfZxx-f13--W9798mrn28be6nK-ubW7ubK3trL36esZdm0smWyNNy3faTslP5neZ3P_LGv0Hvayc2zES4XvYambya6DRydFgkozLOfSGV-i3jvPRg6iduFglIvsGLWLLdpLrDH_eCG7jEXLfHrFuv0Xly7erA |
| linkProvider | EBSCOhost |
| openUrl | ctx_ver=Z39.88-2004&ctx_enc=info%3Aofi%2Fenc%3AUTF-8&rfr_id=info%3Asid%2Fsummon.serialssolutions.com&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=%E5%A4%9A%E4%BB%BB%E5%8A%A1%E5%AD%A6%E4%B9%A0%E5%9C%A8%E5%88%86%E5%AD%90%E6%80%A7%E8%B4%A8%E9%A2%84%E6%B5%8B%E4%B8%AD%E7%9A%84%E5%AF%B9%E6%AF%94%E7%A0%94%E7%A9%B6&rft.jtitle=%E5%8C%96%E5%AD%A6%E7%89%A9%E7%90%86%E5%AD%A6%E6%8A%A5%EF%BC%88%E8%8B%B1%E6%96%87%E7%89%88%EF%BC%89&rft.au=%E9%9F%A9%E8%B6%85&rft.au=%E7%8E%8B%E7%9A%93&rft.au=%E6%9C%B1%E5%81%A5%E4%BF%9D&rft.au=%E5%88%98%E6%B7%87&rft.date=2023&rft.pub=%E4%B8%AD%E5%9B%BD%E7%A7%91%E5%AD%A6%E6%8A%80%E6%9C%AF%E5%A4%A7%E5%AD%A6%E7%89%A9%E7%90%86%E5%AD%A6%E9%99%A2%2C%E5%90%88%E8%82%A5+230026%25%E4%B8%AD%E5%9B%BD%E7%A7%91%E5%AD%A6%E6%8A%80%E6%9C%AF%E5%A4%A7%E5%AD%A6%E8%AE%A1%E7%AE%97%E6%9C%BA%E7%A7%91%E5%AD%A6%E4%B8%8E%E6%8A%80%E6%9C%AF%E5%AD%A6%E9%99%A2%2C%E5%A4%A7%E6%95%B0%E6%8D%AE%E5%88%86%E6%9E%90%E4%B8%8E%E5%BA%94%E7%94%A8%E5%AE%89%E5%BE%BD%E7%9C%81%E9%87%8D%E7%82%B9%E5%AE%9E%E9%AA%8C%E5%AE%A4%2C%E5%90%88%E8%82%A5+230026&rft.issn=1674-0068&rft.volume=36&rft.issue=4&rft.spage=443&rft.epage=452&rft_id=info:doi/10.1063%2F1674-0068%2Fcjcp2203055&rft.externalDocID=hxwlxb202304009 |
| thumbnail_s | http://utb.summon.serialssolutions.com/2.0.0/image/custom?url=http%3A%2F%2Fwww.wanfangdata.com.cn%2Fimages%2FPeriodicalImages%2Fhxwlxb%2Fhxwlxb.jpg |