full-FORCE: A target-based method for training recurrent networks.

full-FORCE: A target-based method for training recurrent networks.
复制标题

DOI:
10.1371/journal.pone.0191527
复制
发表时间:
2018
期刊:
影响因子:
3.7
通讯作者:
Abbott LF
Abbott LF
中科院分区:
综合性期刊3区
文献类型:
--
作者:
DePasquale B;Cueva CJ;Rajan K;Escola GS;Abbott LF

文献摘要

参考文献

被引文献

相似文献

训练的递归网络是建模动态神经计算的强大工具。我们提出了一种基于目标的方法,用于修改递归网络的全连接矩阵,以训练它执行涉及时间复杂的输入/输出变换的任务。该方法在训练期间引入第二网络,以提供对执行任务有用的合适的“目标”动态。因为它利用了完整的递归连接,所以该方法产生的网络比传统的最小二乘(FORCE)方法具有更少的神经元和更高的噪声鲁棒性。此外,我们还展示了如何将额外的输入信号引入目标生成网络,作为任务提示,极大地扩展了可以学习的任务范围,并提供了对经过训练的任务执行网络的动态复杂性和性质的控制。
Trained recurrent networks are powerful tools for modeling dynamic neural computations. We present a target-based method for modifying the full connectivity matrix of a recurrent network to train it to perform tasks involving temporally complex input/output transformations. The method introduces a second network during training to provide suitable “target” dynamics useful for performing the task. Because it exploits the full recurrent connectivity, the method produces networks that perform tasks with fewer neurons and greater noise robustness than traditional least-squares (FORCE) approaches. In addition, we show how introducing additional input signals into the target-generating network, which act as task hints, greatly extends the range of tasks that can be learned and provides control over the complexity and nature of the dynamics of the trained, task-performing network.
DOI: 10.1038/nn.3405
发表时间: 2013-07
影响因子: 25
作者:
Laje, Rodrigo;Buonomano, Dean V.
通讯作者: Buonomano, Dean V.
DOI: 10.1038/nn.4042
发表时间: 2015-07
影响因子: 25
作者:
Sussillo, David;Churchland, Mark M.;Kaufman, Matthew T.;Shenoy, Krishna V.
通讯作者: Shenoy, Krishna V.
DOI: 10.1371/journal.pone.0037372
发表时间: 2012
期刊: PloS one
影响因子: 3.7
作者:
Sussillo D;Abbott LF
通讯作者: Abbott LF
DOI: 10.1038/20939
发表时间: 1999-06-03
期刊: NATURE
影响因子: 64.8
作者:
Romo, R;Brody, CD;Lemus, L
通讯作者: Lemus, L
DOI: 10.1038/nn.4241
发表时间: 2016-03
影响因子: 25
作者:
Abbott LF;DePasquale B;Memmesheimer RM
通讯作者: Memmesheimer RM