连接任务结构和神经网络动态。
Linking task structure and neural network dynamics.
机构信息
Icahn School of Medicine at Mount Sinai, New York, NY, USA.
Department of Neurobiology, Harvard Medical School, Boston, MA, USA.
出版信息
Nat Neurosci. 2022 Jun;25(6):679-681. doi: 10.1038/s41593-022-01090-w.
Abstract
The solutions neural networks find to solve a task are often inscrutable. We have had little insight into why particular structure emerges in a network. By reverse-engineering neural networks from dynamical principles, Dubreuil & Valente et. al. reveal how neural population structure enables computational flexibility.
摘要
神经网络为解决任务而找到的解决方案通常是难以理解的。我们对为什么网络中会出现特定的结构知之甚少。通过从动力学原理反向设计神经网络，Dubreuil 和 Valente 等人揭示了神经群体结构如何使计算具有灵活性。
相似文献
1
Linking task structure and neural network dynamics.连接任务结构和神经网络动态。
Nat Neurosci. 2022 Jun;25(6):679-681. doi: 10.1038/s41593-022-01090-w.
2
Engineering recurrent neural networks from task-relevant manifolds and dynamics.从任务相关流形和动力学中设计递归神经网络。
PLoS Comput Biol. 2020 Aug 12;16(8):e1008128. doi: 10.1371/journal.pcbi.1008128. eCollection 2020 Aug.
3
OCLSTM: Optimized convolutional and long short-term memory neural network model for protein secondary structure prediction.OCLSTM：用于蛋白质二级结构预测的优化卷积和长短期记忆神经网络模型。
PLoS One. 2021 Feb 3;16(2):e0245982. doi: 10.1371/journal.pone.0245982. eCollection 2021.
4
A robust deep neural network for denoising task-based fMRI data: An application to working memory and episodic memory.用于去噪任务态 fMRI 数据的稳健深度神经网络：在工作记忆和情景记忆中的应用。
Med Image Anal. 2020 Feb;60:101622. doi: 10.1016/j.media.2019.101622. Epub 2019 Nov 26.
5
Prediction for Future Yaw Rate Values of Vehicles Using Long Short-Term Memory Network.利用长短时记忆网络预测车辆未来偏航率值。
Sensors (Basel). 2023 Jun 17;23(12):5670. doi: 10.3390/s23125670.
6
Multiplexing working memory and time in the trajectories of neural networks.在神经网络的轨迹中实现工作记忆和时间的复用。
Nat Hum Behav. 2023 Jul;7(7):1170-1184. doi: 10.1038/s41562-023-01592-y. Epub 2023 Apr 20.
7
Long short-term memory neural network for directly inverse design of nanofin metasurface.长短期记忆神经网络直接用于纳米 fin 超表面的逆设计。
Opt Lett. 2022 Jul 1;47(13):3239-3242. doi: 10.1364/OL.458453.
8
Trained recurrent neural networks develop phase-locked limit cycles in a working memory task.经过训练的循环神经网络在工作记忆任务中发展出锁相的极限环。
PLoS Comput Biol. 2024 Feb 5;20(2):e1011852. doi: 10.1371/journal.pcbi.1011852. eCollection 2024 Feb.
9
Effect of recurrent infomax on the information processing capability of input-driven recurrent neural networks.循环信息最大化对输入驱动循环神经网络信息处理能力的影响。
Neurosci Res. 2020 Jul;156:225-233. doi: 10.1016/j.neures.2020.02.001. Epub 2020 Feb 14.
10
Adaptation supports short-term memory in a visual change detection task.适应支持视觉变化检测任务中的短期记忆。
PLoS Comput Biol. 2021 Sep 17;17(9):e1009246. doi: 10.1371/journal.pcbi.1009246. eCollection 2021 Sep.
引用本文的文献
1
Active filtering: a predictive function of recurrent circuits of sensory cortex.主动滤波：感觉皮层循环回路的一种预测功能。
ArXiv. 2025 Jan 17:arXiv:2501.10521v1.
2
Bidirectional mechanisms and emerging strategies for implantable bioelectronic interfaces.可植入生物电子接口的双向机制与新兴策略
Bioact Mater. 2025 Jun 19;52:634-667. doi: 10.1016/j.bioactmat.2025.06.014. eCollection 2025 Oct.
3
Rapid context inference in a thalamocortical model using recurrent neural networks.使用递归神经网络进行丘脑皮质模型中的快速上下文推断。
Nat Commun. 2024 Sep 27;15(1):8275. doi: 10.1038/s41467-024-52289-3.
4
A rubric for human-like agents and NeuroAI.类人代理和神经 AI 的评估标准。
Philos Trans R Soc Lond B Biol Sci. 2023 Jan 30;378(1869):20210446. doi: 10.1098/rstb.2021.0446. Epub 2022 Dec 13.
5
A highly selective response to food in human visual cortex revealed by hypothesis-free voxel decomposition.基于无假设体素分解的人类视觉皮层对食物的高度选择性反应。
Curr Biol. 2022 Oct 10;32(19):4159-4171.e9. doi: 10.1016/j.cub.2022.08.009. Epub 2022 Aug 25.
本文引用的文献
1
Shaping Dynamics With Multiple Populations in Low-Rank Recurrent Networks.低秩递归网络中多群体的动态塑造。
Neural Comput. 2021 May 13;33(6):1572-1615. doi: 10.1162/neco_a_01381.
2
Computation Through Neural Population Dynamics.通过神经群体动力学进行计算。
Annu Rev Neurosci. 2020 Jul 8;43:249-275. doi: 10.1146/annurev-neuro-092619-094115.
3
How to study the neural mechanisms of multiple tasks.如何研究多任务的神经机制。
Curr Opin Behav Sci. 2019 Oct;29:134-143. doi: 10.1016/j.cobeha.2019.07.001. Epub 2019 Sep 9.
4
Frontal cortex neuron types categorically encode single decision variables.前额皮质神经元类型明确地对单一决策变量进行编码。
Nature. 2019 Dec;576(7787):446-451. doi: 10.1038/s41586-019-1816-9. Epub 2019 Dec 4.
5
Reinforcement Learning, Fast and Slow.强化学习：快与慢。
Trends Cogn Sci. 2019 May;23(5):408-422. doi: 10.1016/j.tics.2019.02.006. Epub 2019 Apr 16.
6
Task representations in neural networks trained to perform many cognitive tasks.神经网络中执行多项认知任务的任务表示。
Nat Neurosci. 2019 Feb;22(2):297-306. doi: 10.1038/s41593-018-0310-2. Epub 2019 Jan 14.
7
Motor primitives in space and time via targeted gain modulation in cortical networks.通过靶向增益调制在皮质网络中实现空间和时间上的运动基元。
Nat Neurosci. 2018 Dec;21(12):1774-1783. doi: 10.1038/s41593-018-0276-0. Epub 2018 Nov 26.
8
Cortical population activity within a preserved neural manifold underlies multiple motor behaviors.皮层群体活动在一个保留的神经流形下产生多种运动行为。
Nat Commun. 2018 Oct 12;9(1):4233. doi: 10.1038/s41467-018-06560-z.
9
Linking Connectivity, Dynamics, and Computations in Low-Rank Recurrent Neural Networks.在低秩递归神经网络中连接连通性、动态和计算。
Neuron. 2018 Aug 8;99(3):609-623.e29. doi: 10.1016/j.neuron.2018.07.003. Epub 2018 Jul 26.
10
Computing by Robust Transience: How the Fronto-Parietal Network Performs Sequential, Category-Based Decisions.基于稳健瞬态的计算：额顶叶网络如何执行基于类别的顺序决策。
Neuron. 2017 Mar 22;93(6):1504-1517.e4. doi: 10.1016/j.neuron.2017.03.002.
Zotero 插件