Recurrent neural networks with transient trajectory explain working memory encoding mechanisms
Abstract Whether working memory (WM) is encoded by persistent activity using attractors or by dynamic activity using transient trajectories has been debated for decades in both experimental and modeling studies, and a consensus has not been reached. Even though many recurrent neural networks (RNNs)...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-01-01
|
| Series: | Communications Biology |
| Online Access: | https://doi.org/10.1038/s42003-024-07282-3 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1832571360508903424 |
|---|---|
| author | Chenghao Liu Shuncheng Jia Hongxing Liu Xuanle Zhao Chengyu T. Li Bo Xu Tielin Zhang |
| author_facet | Chenghao Liu Shuncheng Jia Hongxing Liu Xuanle Zhao Chengyu T. Li Bo Xu Tielin Zhang |
| author_sort | Chenghao Liu |
| collection | DOAJ |
| description | Abstract Whether working memory (WM) is encoded by persistent activity using attractors or by dynamic activity using transient trajectories has been debated for decades in both experimental and modeling studies, and a consensus has not been reached. Even though many recurrent neural networks (RNNs) have been proposed to simulate WM, most networks are designed to match respective experimental observations and show either transient or persistent activities. Those few which consider networks with both activity patterns have not attempted to directly compare their memory capabilities. In this study, we build transient-trajectory-based RNNs (TRNNs) and compare them to vanilla RNNs with more persistent activities. The TRNN incorporates biologically plausible modifications, including self-inhibition, sparse connection and hierarchical topology. Besides activity patterns resembling animal recordings and retained versatility to variable encoding time, TRNNs show better performance in delayed choice and spatial memory reinforcement learning tasks. Therefore, this study provides evidence supporting the transient activity theory to explain the WM mechanism from the model designing point of view. |
| format | Article |
| id | doaj-art-22724d0e77064c45b1b7037293924247 |
| institution | Kabale University |
| issn | 2399-3642 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Nature Portfolio |
| record_format | Article |
| series | Communications Biology |
| spelling | doaj-art-22724d0e77064c45b1b70372939242472025-02-02T12:37:21ZengNature PortfolioCommunications Biology2399-36422025-01-018111310.1038/s42003-024-07282-3Recurrent neural networks with transient trajectory explain working memory encoding mechanismsChenghao Liu0Shuncheng Jia1Hongxing Liu2Xuanle Zhao3Chengyu T. Li4Bo Xu5Tielin Zhang6Institute of Automation, Chinese Academy of SciencesInstitute of Automation, Chinese Academy of SciencesInstitute of Automation, Chinese Academy of SciencesInstitute of Automation, Chinese Academy of SciencesLingang LaboratoryInstitute of Automation, Chinese Academy of SciencesInstitute of Automation, Chinese Academy of SciencesAbstract Whether working memory (WM) is encoded by persistent activity using attractors or by dynamic activity using transient trajectories has been debated for decades in both experimental and modeling studies, and a consensus has not been reached. Even though many recurrent neural networks (RNNs) have been proposed to simulate WM, most networks are designed to match respective experimental observations and show either transient or persistent activities. Those few which consider networks with both activity patterns have not attempted to directly compare their memory capabilities. In this study, we build transient-trajectory-based RNNs (TRNNs) and compare them to vanilla RNNs with more persistent activities. The TRNN incorporates biologically plausible modifications, including self-inhibition, sparse connection and hierarchical topology. Besides activity patterns resembling animal recordings and retained versatility to variable encoding time, TRNNs show better performance in delayed choice and spatial memory reinforcement learning tasks. Therefore, this study provides evidence supporting the transient activity theory to explain the WM mechanism from the model designing point of view.https://doi.org/10.1038/s42003-024-07282-3 |
| spellingShingle | Chenghao Liu Shuncheng Jia Hongxing Liu Xuanle Zhao Chengyu T. Li Bo Xu Tielin Zhang Recurrent neural networks with transient trajectory explain working memory encoding mechanisms Communications Biology |
| title | Recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| title_full | Recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| title_fullStr | Recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| title_full_unstemmed | Recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| title_short | Recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| title_sort | recurrent neural networks with transient trajectory explain working memory encoding mechanisms |
| url | https://doi.org/10.1038/s42003-024-07282-3 |
| work_keys_str_mv | AT chenghaoliu recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT shunchengjia recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT hongxingliu recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT xuanlezhao recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT chengyutli recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT boxu recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms AT tielinzhang recurrentneuralnetworkswithtransienttrajectoryexplainworkingmemoryencodingmechanisms |