PsychRNN: An Accessible and Flexible Python']Python Package for Training Recurrent Neural Network Models on Cognitive Tasks

被引:14
作者
Ehrlich, Daniel B. [1 ]
Stone, Jasmine T. [2 ]
Brandfonbrener, David [2 ,3 ]
Atanasov, Alexander [4 ,5 ]
Murray, John D. [1 ,4 ,6 ]
机构
[1] Yale Univ, Interdept Neurosci Program, New Haven, CT 06520 USA
[2] Yale Univ, Dept Comp Sci, POB 2158, New Haven, CT 06520 USA
[3] NYU, Dept Comp Sci, New York, NY 10012 USA
[4] Yale Univ, Dept Phys, New Haven, CT 06511 USA
[5] Harvard Univ, Dept Phys, Cambridge, MA 02138 USA
[6] Yale Sch Med, Dept Psychiat, New Haven, CT 06511 USA
基金
美国国家卫生研究院;
关键词
cognitive task; computational model; deep learning; recurrent neural network; training; FRAMEWORK;
D O I
10.1523/ENEURO.0427-20.2020
中图分类号
Q189 [神经科学];
学科分类号
071006 ;
摘要
Example workflow for using PsychRNN. First, the task of interest is defined, and a recurrent neural network (RNN) model is trained to perform the task, optionally with neurobiologically informed constraints on the network. After the network is trained, the researchers can investigate network properties including the synaptic connectivity patterns and the dynamics of neural population activity during task execution, and other studies, e.g., those on perturbations, can be explored. The dotted line shows the possible repetition of this cycle with one network, which allows investigation of training effects of task shaping, or curriculum learning, for closed-loop training of the network on a progression of tasks. Task-trained artificial recurrent neural networks (RNNs) provide a computational modeling framework of increasing interest and application in computational, systems, and cognitive neuroscience. RNNs can be trained, using deep-learning methods, to perform cognitive tasks used in animal and human experiments and can be studied to investigate potential neural representations and circuit mechanisms underlying cognitive computations and behavior. Widespread application of these approaches within neuroscience has been limited by technical barriers in use of deep-learning software packages to train network models. Here, we introduce PsychRNN, an accessible, flexible, and extensible Python package for training RNNs on cognitive tasks. Our package is designed for accessibility, for researchers to define tasks and train RNN models using only Python and NumPy, without requiring knowledge of deep-learning software. The training backend is based on TensorFlow and is readily extensible for researchers with Tensor Flow knowledge to develop projects with additional customization. PsychRNN implements a number of specialized features to support applications in systems and cognitive neuroscience. Users can impose neurobiologically relevant constraints on synaptic connectivity patterns. Furthermore, specification of cognitive tasks has a modular structure, which facilitates parametric variation of task demands to examine their impact on model solutions. PsychRNN also enables task shaping during training, or curriculum learning, in which tasks are adjusted in closed-loop based on performance. Shaping is ubiquitous in training of animals in cognitive tasks, and PsychRNN allows investigation of how shaping trajectories impact learning and model solutions. Overall, the PsychRNN framework facilitates application of trained RNNs in neuroscience research.
引用
收藏
页码:1 / 11
页数:11
相关论文
共 32 条
[1]   Recurrent neural networks as versatile tools of neuroscience research [J].
Barak, Omri .
CURRENT OPINION IN NEUROBIOLOGY, 2017, 46 :1-6
[2]   Learning Deep Architectures for AI [J].
Bengio, Yoshua .
FOUNDATIONS AND TRENDS IN MACHINE LEARNING, 2009, 2 (01) :1-127
[3]   Standardized automated training of rhesus monkeys for neuroscience research in their housing environment [J].
Berger, M. ;
Calapai, A. ;
Stephan, V. ;
Niessing, M. ;
Burchardt, L. ;
Gail, A. ;
Treue, S. .
JOURNAL OF NEUROPHYSIOLOGY, 2018, 119 (03) :796-807
[4]   Deep Reinforcement Learning and Its Neuroscientific Implications [J].
Botvinick, Matthew ;
Wang, Jane X. ;
Dabney, Will ;
Miller, Kevin J. ;
Kurth-Nelson, Zeb .
NEURON, 2020, 107 (04) :603-616
[5]   Dynamic Control of Response Criterion in Premotor Cortex during Perceptual Detection under Temporal Uncertainty [J].
Carnevale, Federico ;
de Lafuente, Victor ;
Romo, Ranulfo ;
Barak, Omri ;
Parga, Nestor .
NEURON, 2015, 86 (04) :1067-1077
[6]  
Chollet F., 2015, KERAS 20 COMPUTER SO
[7]   Experience-dependent representation of visual categories in parietal cortex [J].
Freedman, David J. ;
Assad, John A. .
NATURE, 2006, 443 (7107) :85-88
[8]  
Glorot X., 2010, Proceedings of the thirteenth international conference on artificial intelligence and statistics, P249, DOI DOI 10.1109/LGRS.2016.2565705
[9]  
Graves A, 2012, STUD COMPUT INTELL, V385, P1, DOI [10.1162/neco.1997.9.1.1, 10.1007/978-3-642-24797-2]
[10]   Deep Neural Networks: A New Framework for Modeling Biological Vision and Brain Information Processing [J].
Kriegeskorte, Nikolaus .
ANNUAL REVIEW OF VISION SCIENCE, VOL 1, 2015, 1 :417-446