Training Excitatory-Inhibitory Recurrent Neural Networks for Cognitive Tasks: A Simple and Flexible Framework
Open accessThe DOI matches its Crossref record; 91 of the 91 references it cites were found in a registry or an index.
Abstract
The ability to simultaneously record from large numbers of neurons in behaving animals has ushered in a new era for the study of the neural circuit mechanisms underlying cognitive functions. One promising approach to uncovering the dynamical and computational principles governing population responses is to analyze model recurrent neural networks (RNNs) that have been optimized to perform the same tasks as behaving animals. Because the optimization of network parameters specifies the desired output but not the manner in which to achieve this output, "trained" networks serve as a source of mechanistic hypotheses and a testing ground for data analyses that link neural computation to behavior. Complete access to the activity and connectivity of the circuit, and the ability to manipulate them arbitrarily, make trained networks a convenient proxy for biological circuits and a valuable platform for theoretical investigation. However, existing RNNs lack basic biological features such as the distinction between excitatory and inhibitory units (Dale's principle), which are essential if RNNs are to provide insights into the operation of biological circuits. Moreover, trained networks can achieve the same behavioral performance but differ substantially in their structure and dynamics, highlighting the need for a simple and flexible framework for the exploratory training of RNNs. Here, we describe a framework for gradient descent-based training of excitatory-inhibitory RNNs that can incorporate a variety of biological knowledge. We provide an implementation based on the machine learning library Theano, whose automatic differentiation capabilities facilitate modifications and extensions. We validate this framework by applying it to well-known experimental paradigms such as perceptual decision-making, context-dependent integration, multisensory integration, parametric working memory, and motor sequence generation. Our results demonstrate the wide range of neural activity patterns and behavior that can be modeled, and suggest a unified setting in which diverse cognitive computations and mechanisms can be studied.
Overview
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Integrity signals
Registry
Cites
Statistics
Figures



![Fig 4 Context-dependent integration task. (A) Psychometric function, showing the percentage of trials on which the network chose choice 1 as a function of the signed motion (upper) and signed color (lower) coherence in motion-context (black) and color-context (blue) trials. (B) Average population responses in state space during the stimulus period, projected to the 3-dimensional subspace capturing variance due to choice, motion, and color as in [5]. Only correct trials were included. The task-related axes were obtained through a linear regression analysis. Note that “choice” here has a unit-specific meaning that depends on the preferred choice of the unit as determined by the selectivity index (Eq 30). For both motion (black) and color (blue), coherences increase from light to dark. Upper plots show trials during the motion context, and lower plots show trials during the color context. (C) Normalized responses of four recurrent units during the stimulus period show mixed representation of task variables. Solid lines indicate the preferred choice and dashed lines the nonpreferred choice of each unit. (D) Denoised regression coefficients from the linear regression analysis. By definition, the coefficients for choice are almost exclusively positive.](https://figures.cochat.ai/paper-figures/ca3a2b623c1e4966aef22596dc894038/training-excitatory-inhibitory-recurrent-neural-networks-for-cognitive-tasks-a-figure-4.jpg)





Verification
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The cited source says what the claim attributes to it.
In the citing paper
“and which can be viewed as an approximation to continuous time firing rate models of neural circuits (Song et al., 2016).”
From Remapping in a recurrent neural network model of navigation and context inference (2023) · cites this paper as Song et al., 2016
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Citations
Cites
Hochreiter, S., & Schmidhuber, J. (1997) · Neural Computation · doi:10.1162/neco.1997.9.8.1735 · Cited by 102,318 · cited 1×
“For instance, machine learning applications typically employ a type of RNN known as Long Short-Term Memory (LSTM), which uses multiplicative gates to facilitate learning of long-term dependencies and currently represents one of the most powerful methods for solving sequence-related problems [72].”
MethodReinforcement Learning: An Introduction
Sutton, R. S., & Barto, A. G. (1998) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.1998.712192 · Cited by 27,225 · cited 1×
“Combining RNN training with reinforcement learning methods [75 – 77] will be essential to bringing network training closer to the reward-based manner in which animals are trained.”
MethodReinforcement Learning: An Introduction
(2005) · IEEE Transactions on Neural Networks · doi:10.1109/tnn.2004.842673 · Cited by 24,496
Learning Internal Representations by Error Propagation
Rumelhart, D. E., Hinton, G. E., & Williams, R. J. (1985) · doi:10.21236/ada164453 · Cited by 16,172
Parallel Distributed Processing: Explorations in the Microstructure of Cognition: Foundations
Rumelhart, D. E., McClelland, J. L., & AU (1986) · doi:10.7551/mitpress/5236.001.0001 · Cited by 15,686
Learning long-term dependencies with gradient descent is difficult
Bengio, Y., Simard, P., & Frasconi, P. (1994) · IEEE Transactions on Neural Networks · doi:10.1109/72.279181 · Cited by 8,542 · cited 1×
“Even so, SGD with the objective function given in Eqs 15 and 16 often fails to converge to a solution when the network must learn dependencies between distant time points [60].”
BackgroundOn the difficulty of training recurrent neural networks
Pascanu, R., Mikolov, T., & Bengio, Y. (2012) · arXiv · doi:10.48550/arxiv.1211.5063 · Cited by 4,310 · cited 4×
“To remedy this problem, which is due to some gradient components being too large ( exploding gradients) and some gradient components being too small ( vanishing gradients), we follow [35] in making two modifications.”
BackgroundMethodInfluential(2013) · Cited by 3,740
Spiking Neuron Models: Single Neurons, Populations, Plasticity
Gerstner, W., & Kistler, W. M. (2002) · Cambridge University Press eBooks · doi:10.1017/cbo9780511815706 · Cited by 3,080
Interneurons of the neocortical inhibitory system
Markram, H., Toledo-Rodriguez, M., Wang, Y., Gupta, A., Silberberg, G., & Wu, C. (2004) · Nature Reviews Neuroscience · doi:10.1038/nrn1519 · Cited by 2,954 · cited 2×
“Local cortical circuits have distance [48], layer [26, 49, 50], and cell-type [23, 25, 27, 51] dependent patterns of connectivity and different overall levels of sparseness for excitatory to excitatory, inhibitory to excitatory, excitatory to inhibitory, and inhibitory to inhibitory connections [52, 53].”
BackgroundRumelhart, D. E., & McClelland, J. L. (1986) · The MIT Press eBooks · Cited by 2,906
Mnemonic coding of visual space in the monkey's dorsolateral prefrontal cortex
Funahashi, S., Bruce, C. J., & Goldman-Rakic, P. S. (1989) · Journal of Neurophysiology · doi:10.1152/jn.1989.61.2.331 · Cited by 2,685
Cohen, J. D., Dunbar, K., & McClelland, J. L. (1990) · Psychological Review · doi:10.1037//0033-295x.97.3.332 · Cited by 2,139 · cited 1×
“Answering this question is essential if RNNs are to provide insights into the operation of the brain at the level of neural circuits [18], and extends the classical connectionist approach [19, 20].”
BackgroundSpiking Neuron Models: Single Neurons, Populations, Plasticity
Wulfram Gerstner, B., & Kistler, W. M. (2002) · Cited by 2,028
Performance-optimized hierarchical models predict neural responses in higher visual cortex
Yamins, D. L. K., Hong, H., Cadieu, C. F., Solomon, E. A., Seibert, D., & DiCarlo, J. J. (2014) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1403112111 · Cited by 1,924 · cited 1×
“Complete access to the activity and connectivity of the circuit, and the ability to manipulate them in arbitrary ways, make trained networks a convenient proxy for biological circuits and a valuable platform for theoretical investigation [12, 16, 17].”
BackgroundContext-dependent computation by recurrent dynamics in prefrontal cortex
Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013) · Nature · doi:10.1038/nature12742 · Cited by 1,847 · cited 10×
“A growing recognition of the importance of studying population-level responses is reflected in the increasing number of studies that use large data-sets of simultaneously or sequentially recorded neurons to infer neural circuit mechanisms [5 – 9].”
BackgroundMethodResultInfluentialThe importance of mixed selectivity in complex cognitive tasks
Rigotti, M., Barak, O., Warden, M. R., Wang, X.-J., Daw, N. D., Miller, E. K., & Fusi, S. (2013) · Nature · doi:10.1038/nature12160 · Cited by 1,639
Self-citationNeural population dynamics during reaching
Churchland, M. M., Cunningham, J. P., Kaufman, M. T., Foster, J. D., Nuyujukian, P., Ryu, S. I., & Shenoy, K. V. (2012) · Nature · doi:10.1038/nature11129 · Cited by 1,626
Roitman, J. D., & Shadlen, M. N. (2002) · The Journal of Neuroscience · doi:10.1523/jneurosci.22-21-09475.2002 · Cited by 1,589 · cited 3×
“To control the speed-accuracy tradeoff, the target outputs during training did not require the network to commit to a decision immediately but instead after a short delay [62]; the delay determines the cost incurred for answering early but incorrectly versus correctly but at a later time.”
BackgroundInfluentialHighly nonrandom features of synaptic connectivity in local cortical circuits
Song, S., Sjöström, P. J., Reigl, M., Nelson, S., & Chklovskii, D. B. (2005) · PLoS Biology · doi:10.1371/journal.pbio.0030068 · Cited by 1,512
Pfeffer, C. K., Xue, M., He, M., Huang, Z. J., & Scanziani, M. (2013) · Nature Neuroscience · doi:10.1038/nn.3446 · Cited by 1,353 · cited 2×
“…of excitatory and inhibitory neuronal morphology and physiology, their layer and type-specific distribution and connectivity [49, 50], and domain-specific (e.g., dendritic versus somatic) targeting of excitatory pyramidal cells by interneurons [23, 25, 27, 51] in the brain are taken into account.”
BackgroundDimensionality reduction for large-scale neural recordings
Cunningham, J. P., & Yu, B. M. (2014) · Nature Neuroscience · doi:10.1038/nn.3776 · Cited by 1,219
Cholinergic and inhibitory synapses in a pathway from motor‐axon collaterals to motoneurones
Eccles, J. C., Fatt, P., & Koketsu, K. (1954) · The Journal of Physiology · doi:10.1113/jphysiol.1954.sp005226 · Cited by 1,193
Zipser, D., & Andersen, R. A. (1988) · Nature · doi:10.1038/331679a0 · Cited by 1,187 · cited 1×
“Answering this question is essential if RNNs are to provide insights into the operation of the brain at the level of neural circuits [18], and extends the classical connectionist approach [19, 20].”
BackgroundNeuronal correlates of a perceptual decision
Newsome, W. T., Britten, K. H., & Movshon, J. A. (1989) · Nature · doi:10.1038/341052a0 · Cited by 1,177 · cited 2×
“…we present the results of applying the training framework to well-known experimental paradigms in systems neuroscience: perceptual decision-making [61 – 63], context-dependent integration [5], multisensory integration [64], parametric working memory [34, 65], and eye-movement sequence generation…”
MethodThe Asynchronous State in Cortical Circuits
Renart, A., de la Rocha, J., Bartho, P., Hollender, L., Parga, N., Reyes, A., & Harris, K. D. (2010) · Science · doi:10.1126/science.1179850 · Cited by 1,106 · cited 1×
“, faster inhibitory units), correlated noise [73], and other types of nonlinearities besides simple rectification (e.”
Generating Coherent Patterns of Activity from Chaotic Neural Networks
Sussillo, D., & Abbott, L. F. (2009) · Neuron · doi:10.1016/j.neuron.2009.07.018 · Cited by 1,092 · cited 2×
“…it is convenient to initialize the recurrent weight matrix as W rec ¼ r W rec 0 , where W rec 0 is formed by setting a fraction p , 0 p (cid:4) 1, of elements to nonzero values drawn from a Gaussian distribution with mean 0 and variance ( pN ) − 1 , and the remaining fraction 1 − p to zero [31].”
Theano: A CPU and GPU Math Compiler in Python
Bergstra, J., Breuleux, O., Bastien, F., Lamblin, P., Pascanu, R., Desjardins, G., Turian, J., Warde-Farley, D., & Bengio, Y. (2010) · Proceedings of the Python in Science Conference · doi:10.25080/majora-92bf1922-003 · Cited by 1,052 · cited 2×
“We provide an implementation of this framework based on the Python machine learning library Theano [41, 42], whose automatic differentiation capabilities facilitate modifications and extensions.”
MethodA Recurrent Network Mechanism of Time Integration in Perceptual Decisions
Wong, K.-F., & Wang, X.-J. (2006) · The Journal of Neuroscience · doi:10.1523/jneurosci.3733-05.2006 · Cited by 1,029 · cited 1×
“In the case of the network with separate excitatory and inhibitory populations (Fig 3B), units with different d ′ interact primarily through inhibitory units [67].”
BackgroundSelf-citationTheano: new features and speed improvements
Bastien, F., Lamblin, P., Pascanu, R., Bergstra, J., Goodfellow, I., Bergeron, A., Bouchard, N., Warde-Farley, D., & Bengio, Y. (2012) · arXiv · doi:10.48550/arxiv.1211.5590 · Cited by 1,005
Approximation of dynamical systems by continuous time recurrent neural networks
Funahashi, K.-I., & Nakamura, Y. (1993) · Neural Networks · doi:10.1016/s0893-6080(05)80125-x · Cited by 980
Principles of connectivity among morphologically defined cell types in adult neocortex
Jiang, X., Shen, S., Cadwell, C. R., Berens, P., Sinz, F., Ecker, A. S., Patel, S., & Tolias, A. S. (2015) · Science · doi:10.1126/science.aac9462 · Cited by 908 · cited 3×
“…of excitatory and inhibitory neuronal morphology and physiology, their layer and type-specific distribution and connectivity [49, 50], and domain-specific (e.g., dendritic versus somatic) targeting of excitatory pyramidal cells by interneurons [23, 25, 27, 51] in the brain are taken into account.”
BackgroundA Quantitative Map of the Circuit of Cat Primary Visual Cortex
Binzegger, T., Douglas, R. J., & Martin, K. A. C. (2004) · The Journal of Neuroscience · doi:10.1523/jneurosci.1400-04.2004 · Cited by 878 · cited 2×
“…the full diversity of excitatory and inhibitory neuronal morphology and physiology, their layer and type-specific distribution and connectivity [49, 50], and domain-specific (e.g., dendritic versus somatic) targeting of excitatory pyramidal cells by interneurons [23, 25, 27, 51] in the brain are…”
BackgroundSynaptic basis of cortical persistent activity: the importance of NMDA receptors to working memory
Wang, X.-J. (1999) · The Journal of Neuroscience · doi:10.1523/jneurosci.19-21-09587.1999 · Cited by 859 · cited 1×
“One of the most important — and therefore one of the most widely studied — cognitive functions is working memory, the ability to maintain and manipulate information for several seconds during the planning and execution of a task [69, 70].”
BackgroundSelf-citationA Weighted and Directed Interareal Connectivity Matrix for Macaque Cerebral Cortex
Markov, N. T., Ercsey-Ravasz, M. M., Ribeiro Gomes, A. R., Lamy, C., Magrou, L., Vezoli, J., Misery, P., Falchier, A., Quilodran, R., Gariel, M. A., et al. (2014) · Cerebral Cortex · doi:10.1093/cercor/bhs270 · Cited by 857
On the Number of Linear Regions of Deep Neural Networks
Montúfar, G., Pascanu, R., Cho, K., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1402.1869 · Cited by 840 · cited 1×
“This toggles the network between different linear maps, thereby endowing the network with the capacity for more complex computations than would be possible with a single linear network [44, 45].”
BackgroundIdentifying and attacking the saddle point problem in high-dimensional non-convex optimization
Dauphin, Y., Pascanu, R., Gulcehre, C., Cho, K., Ganguli, S., & Bengio, Y. (2014) · arXiv · doi:10.48550/arxiv.1406.2572 · Cited by 827 · cited 1×
“…to a satisfactory solution when the objective function is a highly complicated function of the parameters by sto-chastically sampling the gradient and thereby escaping saddle points [59] or poor local minima, while still performing an averaged form of gradient descent over many stochastic updates.”
BackgroundNeuronal correlates of parametric working memory in the prefrontal cortex
Romo, R., Brody, C. D., Hernández, A., & Lemus, L. (1999) · Nature · doi:10.1038/20939 · Cited by 822 · cited 2×
“…present the results of applying the training framework to well-known experimental paradigms in systems neuroscience: perceptual decision-making [61 – 63], context-dependent integration [5], multisensory integration [64], parametric working memory [34, 65], and eye-movement sequence generation [66].”
MethodFrom the neuron doctrine to neural networks
Yuste, R. (2015) · Nature Reviews Neuroscience · doi:10.1038/nrn3962 · Cited by 754 · cited 1×
“…responses can reveal a great deal about the neural mechanisms underlying various sensory, motor, and cognitive functions, neural mechanisms often involve the coordinated activity of many neurons whose complex individual dynamics are not easily explained by tuning to experimental parameters [1 – 4].”
BackgroundOn the Computational Power of Neural Nets
Siegelmann, H. T., & Sontag, E. D. (1995) · Journal of Computer and System Sciences · doi:10.1006/jcss.1995.1013 · Cited by 692
Spike Timing-Dependent Plasticity: From Synapse to Perception
Dan, Y., & Poo, M.-M. (2006) · Physiological Reviews · doi:10.1152/physrev.00030.2005 · Cited by 675 · cited 2×
“Although none of the learning methods discussed here can at present be considered biological, recent work also suggests that spike-timing dependent plasticity (STDP) [38], which is believed to be a basic rule governing synaptic weight changes in the brain, may correspond to a form of SGD [39, 40].”
BackgroundSolving the Distal Reward Problem through Linkage of STDP and Dopamine Signaling
Izhikevich, E. M. (2007) · Cerebral Cortex · doi:10.1093/cercor/bhl152 · Cited by 657
Rats and Humans Can Optimally Accumulate Evidence for Decision-Making
Brunton, B. W., Botvinick, M. M., & Brody, C. D. (2013) · Science · doi:10.1126/science.1233912 · Cited by 620 · cited 1×
“However, the two sources of noise can be biologically and conceptually quite different [47], and for this reason it is helpful to separate the two types of noise in our formulation.”
BackgroundA neural network that finds a naturalistic solution for the production of muscle activity
Sussillo, D., Churchland, M. M., Kaufman, M. T., & Shenoy, K. V. (2015) · Nature Neuroscience · doi:10.1038/nn.4042 · Cited by 598 · cited 6×
“SGD is conceptually simple without sacrificing performance [36, 37] and is particularly advantageous in the present context for the following reasons: Unlike FORCE and like HF, SGD allows us to more easily formulate the problem of training an RNN as one of minimizing an objective function that can be modified to induc…”
BackgroundMethodResultInfluentialHow advances in neural recording affect data analysis
Stevenson, I. H., & Kording, K. P. (2011) · Nature Neuroscience · doi:10.1038/nn.2731 · Cited by 582 · cited 1×
“At the same time, the novel challenges posed by high-dimensional neural data have led to the development of new methods for analyzing and modeling such data [10 – 12].”
Opening the black box: low-dimensional dynamics in high-dimensional recurrent neural networks
Sussillo, D., & Barak, O. (2013) · Neural Computation · doi:10.1162/neco_a_00409 · Cited by 575 · cited 1×
“In this regard, although state-space analyses of fixed and “slow” points [83] are illuminating they do not yet explain how the network’s connectivity, combined with the nonlinear activation functions, lead to the observed neural trajectories.”
BackgroundRumelhart, D. E., & McClelland, J. L. (1986) · OSTI OAI (U.S. Department of Energy Office of Scientific and Technical Information) · Cited by 553
Learning Recurrent Neural Networks with Hessian-Free Optimization
Martens, J., & Sutskever, I. (2011) · International Conference on Machine Learning · Cited by 525
A Universal Model for Spike-Frequency Adaptation
Benda, J., & Herz, A. V. M. (2003) · Neural Computation · doi:10.1162/089976603322385063 · Cited by 522 · cited 1×
“Other physiologically relevant phenomena such as bursting, adaptation, and oscillations are currently not captured by our framework, but can be incorporated in the future; adaptation, for example, can be included in phenomenological form appropriate to a rate model [81, 82].”
BackgroundA Simple Way to Initialize Recurrent Networks of Rectified Linear Units
Le, Q. V., Jaitly, N., & Hinton, G. E. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1504.00941 · Cited by 513
Kiani, R., Hanks, T. D., & Shadlen, M. N. (2008) · The Journal of Neuroscience · doi:10.1523/jneurosci.4761-07.2008 · Cited by 500 · cited 4×
“…present the results of applying the training framework to well-known experimental paradigms in systems neuroscience: perceptual decision-making [61 – 63], context-dependent integration [5], multisensory integration [64], parametric working memory [34, 65], and eye-movement sequence generation [66].”
BackgroundMethodInfluentialA category-free neural population supports evolving demands during decision-making
Raposo, D., Kaufman, M. T., & Churchland, A. K. (2014) · Nature Neuroscience · doi:10.1038/nn.3865 · Cited by 489 · cited 2×
“…present the results of applying the training framework to well-known experimental paradigms in systems neuroscience: perceptual decision-making [61 – 63], context-dependent integration [5], multisensory integration [64], parametric working memory [34, 65], and eye-movement sequence generation [66].”
BackgroundMethodA Predictive Network Model of Cerebral Cortical Connectivity Based on a Distance Rule
Ercsey-Ravasz, M., Markov, N. T., Lamy, C., Van Essen, D. C., Knoblauch, K., Toroczkai, Z., & Kennedy, H. (2013) · Neuron · doi:10.1016/j.neuron.2013.07.036 · Cited by 487 · cited 2×
“Moreover, connections from excitatory and inhibitory neurons exhibit different levels of sparseness and specificity, with non-random features in the distribution of connection patterns among neurons both within local circuits [23 – 27] and among cortical areas [28 – 30].”
Advances in optimizing recurrent networks
Bengio, Y., Boulanger-Lewandowski, N., & Pascanu, R. (2013) · 2013 IEEE International Conference on Acoustics, Speech and Signal Processing · doi:10.1109/icassp.2013.6639349 · Cited by 469 · cited 1×
“…remove the major difficulties associated with pure gradient descent training of RNNs. SGD is conceptually simple without sacrificing performance [36, 37] and is particularly advantageous in the pres-ent context for the following reasons: Unlike FORCE and like HF, SGD allows us to more easily…”
BackgroundRobust timing and motor patterns by taming chaos in recurrent neural networks
Laje, R., & Buonomano, D. V. (2013) · Nature Neuroscience · doi:10.1038/nn.3405 · Cited by 454 · cited 1×
“Several different methods have previously been used to train RNNs for cognitive tasks in neuroscience, including first-order reduced and controlled error (FORCE) [7, 31, 32] and Hessian-free (HF) [5, 33, 34].”
MethodCortical information flow during flexible sensorimotor decisions
Siegel, M., Buschman, T. J., & Miller, E. K. (2015) · Science · doi:10.1126/science.aab0551 · Cited by 443
Potjans, T. C., & Diesmann, M. (2014) · Cerebral Cortex · doi:10.1093/cercor/bhs358 · Cited by 440 · cited 1×
“Local cortical circuits have distance [48], layer [26, 49, 50], and cell-type [23, 25, 27, 51] dependent patterns of connectivity and different overall levels of sparseness for excitatory to excitatory, inhibitory to excitatory, excitatory to inhibitory, and inhibitory to inhibitory connections…”
BackgroundThomson, A. M. (2002) · Cerebral Cortex · doi:10.1093/cercor/12.9.936 · Cited by 420 · cited 2×
“…the full diversity of excitatory and inhibitory neuronal morphology and physiology, their layer and type-specific distribution and connectivity [49, 50], and domain-specific (e.g., dendritic versus somatic) targeting of excitatory pyramidal cells by interneurons [23, 25, 27, 51] in the brain…”
BackgroundWang, X.-J., Tegnér, J., Constantinidis, C., & Goldman-Rakic, P. S. (2004) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.0305337101 · Cited by 410 · cited 2×
“…of excitatory and inhibitory neuronal morphology and physiology, their layer and type-specific distribution and connectivity [49, 50], and domain-specific (e.g., dendritic versus somatic) targeting of excitatory pyramidal cells by interneurons [23, 25, 27, 51] in the brain are taken into account.”
BackgroundSelf-citationOn simplicity and complexity in the brave new world of large-scale neuroscience
Gao, P., & Ganguli, S. (2015) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2015.04.003 · Cited by 405 · cited 2×
“At the same time, the novel challenges posed by high-dimensional neural data have led to the development of new methods for analyzing and modeling such data [10 – 12].”
BackgroundBalanced Amplification: A New Mechanism of Selective Amplification of Neural Activity Patterns
Murphy, B. K., & Miller, K. D. (2009) · Neuron · doi:10.1016/j.neuron.2009.02.005 · Cited by 340 · cited 1×
“The analogous constraint that all connection weights from a given unit must have the same sign can have a profound effect on the types of dynamics, such as non-normality [22], that operate in the circuit.”
BackgroundRubin, D. B., Van Hooser, S. D., & Miller, K. D. (2015) · Neuron · doi:10.1016/j.neuron.2014.12.026 · Cited by 331 · cited 1×
“…output units (corresponding to variable thresholds), different time constants for each unit (e.g., faster inhibitory units), correlated noise [73], and other types of nonlinearities besides simple rectification (e.g., supralinear [74] or saturating f-I curves) for either recurrent units or outputs.”
BackgroundThe mathematics of Brownian motion and Johnson noise
Gillespie, D. T. (1996) · American Journal of Physics · doi:10.1119/1.18210 · Cited by 324 · cited 1×
“…form (which we indicate by writing time as a subscript, X t = X ( t (cid:5) Δ t ) for a time-dependent variable X ) in time steps of size Δ t as [46] x where α = Δ t / τ and N (0, 1) are normally distributed random numbers with zero mean and unit variance, sampled independently at every time…”
BackgroundTrain faster, generalize better: Stability of stochastic gradient descent
Hardt, M., Recht, B., & Singer, Y. (2015) · arXiv (Cornell University) · doi:10.48550/arxiv.1509.01240 · Cited by 323
Eigenvalue Spectra of Random Matrices for Neural Networks
Rajan, K., & Abbott, L. F. (2006) · Physical Review Letters · doi:10.1103/physrevlett.97.188104 · Cited by 320 · cited 1×
“The means μ E (excitatory) and μ I (inhibitory) of the gamma distributions are chosen to balance the excitatory and inhibitory inputs to each unit [55], i.e., ∑ j 2 exc | μ j | = ∑ j 2 inh | μ j |, with the overall mean set by the imposed spectral radius ρ .”
BackgroundWorking models of working memory
Barak, O., & Tsodyks, M. (2014) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2013.10.008 · Cited by 278 · cited 1×
“One of the most important — and therefore one of the most widely studied — cognitive functions is working memory, the ability to maintain and manipulate information for several seconds during the planning and execution of a task [69, 70].”
BackgroundAttention-Gated Reinforcement Learning of Internal Representations for Classification
Roelfsema, P. R., & Ooyen, A. V. (2005) · Neural Computation · doi:10.1162/0899766054615699 · Cited by 260 · cited 1×
“Combining RNN training with reinforcement learning methods [75 – 77] will be essential to bringing network training closer to the reward-based manner in which animals are trained.”
MethodNeural circuits as computational dynamical systems
Sussillo, D. (2014) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2014.01.008 · Cited by 247 · cited 1×
“Complete access to the activity and connectivity of the circuit, and the ability to manipulate them in arbitrary ways, make trained networks a convenient proxy for biological circuits and a valuable platform for theoretical investigation [12, 16, 17].”
BackgroundSpatial Profile of Excitatory and Inhibitory Synaptic Connectivity in Mouse Primary Auditory Cortex
Levy, R. B., & Reyes, A. D. (2012) · The Journal of Neuroscience · doi:10.1523/jneurosci.5158-11.2012 · Cited by 246 · cited 1×
“Local cortical circuits have distance [48], layer [26, 49, 50], and cell-type [23, 25, 27, 51] dependent patterns of connectivity and different overall levels of sparseness for excitatory to excitatory, inhibitory to excitatory, excitatory to inhibitory, and inhibitory to inhibitory connections…”
BackgroundReinforcement Learning with Long Short-Term Memory
Bakker, B. (2001) · Cited by 209
Laje, R., & Buonomano, D. V. (2012) · arXiv (Cornell University) · doi:10.48550/arxiv.1210.2104 · Cited by 200 · cited 1×
“Several different methods have previously been used to train RNNs for cognitive tasks in neuroscience, including first-order reduced and controlled error (FORCE) [7, 31, 32] and Hessian-free (HF) [5, 33, 34].”
MethodNeuronal Population Coding of Parametric Working Memory
Barak, O., Tsodyks, M., & Romo, R. (2010) · Journal of Neuroscience · doi:10.1523/jneurosci.1875-10.2010 · Cited by 191 · cited 2×
“…responses can reveal a great deal about the neural mechanisms underlying various sensory, motor, and cognitive functions, neural mechanisms often involve the coordinated activity of many neurons whose complex individual dynamics are not easily explained by tuning to experimental parameters [1 – 4].”
BackgroundMethodA blanket of inhibition: functional inferences from dense inhibitory connectivity
Karnani, M. M., Agetsuma, M., & Yuste, R. (2014) · Current Opinion in Neurobiology · doi:10.1016/j.conb.2013.12.015 · Cited by 188 · cited 2×
“5 from inhibitory units to all other units; these connection probabilities are consistent with what is known for local microcircuits in cortex [52, 53].”
BackgroundRigotti, M. (2010) · Frontiers in Computational Neuroscience · doi:10.3389/fncom.2010.00024 · Cited by 184
The Logic of Inhibitory Connectivity in the Neocortex
Fino, É., Packer, A. M., & Yuste, R. (2012) · The Neuroscientist · doi:10.1177/1073858412456743 · Cited by 178 · cited 2×
“…cortical circuits have distance [48], layer [26, 49, 50], and cell-type [23, 25, 27, 51] dependent patterns of connectivity and different overall levels of sparseness for excitatory to excitatory, inhibitory to excitatory, excitatory to inhibitory, and inhibitory to inhibitory connections [52, 53].”
BackgroundInfluentialFrom fixed points to chaos: Three models of delayed discrimination
Barak, O., Sussillo, D., Romo, R., Tsodyks, M., & Abbott, L. F. (2013) · Progress in Neurobiology · doi:10.1016/j.pneurobio.2013.02.002 · Cited by 171 · cited 5×
“The network ’ s performance on each condition is shown in Fig 7B. Based on the experimental results, we trained the network until the lowest percentage of correct responses in any condition exceeded 85%; for most conditions the performance is much higher [34].”
BackgroundMethodInfluentialSolving the distal reward problem through linkage of STDP and dopamine signaling
Izhikevich, E. M. (2007) · BMC Neuroscience · doi:10.1186/1471-2202-8-s2-s15 · Cited by 163
On the number of response regions of deep feed forward networks with piece-wise linear activations
Pascanu, R., Montufar, G., & Bengio, Y. (2013) · arXiv · doi:10.48550/arxiv.1312.6098 · Cited by 153
Highly nonrandom features of synaptic connectivity in local cortical circuits (vol 3, pg e68, 2005)
Song, S.-F., Sjöström, P. J., Reigl, M., Nelson, S. B., & Chklovskii, D. B. (2005) · UCL Discovery (University College London) · Cited by 147
A recurrent network model of somatosensory parametric working memory in the prefrontal cortex
Miller, P. (2003) · Cerebral Cortex · doi:10.1093/cercor/bhg101 · Cited by 144 · cited 1×
“As a result, giving the network both positively tuned (increasing function of event rate) and negatively tuned (decreasing function of event rate) inputs [68] greatly improved training.”
BackgroundCarnevale, F., de Lafuente, V., Romo, R., Barak, O., & Parga, N. (2015) · Neuron · doi:10.1016/j.neuron.2015.04.014 · Cited by 130 · cited 6×
“Meanwhile, like FORCE and unlike HF, for many tasks SGD can update parameters on a trial-by-trial basis, i.e., in an “ online ” fashion.”
BackgroundMethodInfluentialOn the piecewise analysis of networks of linear threshold neurons
Hahnloser, R. L. T. (1998) · Neural Networks · doi:10.1016/s0893-6080(98)00012-4 · Cited by 124
Spatial embedding of structural similarity in the cerebral cortex
Song, H. F., Kennedy, H., & Wang, X. (2014) · Proceedings of the National Academy of Sciences · doi:10.1073/pnas.1414153111 · Cited by 112 · cited 2×
“Moreover, connections from excitatory and inhibitory neurons exhibit different levels of sparseness and specificity, with non-random features in the distribution of connection patterns among neurons both within local circuits [23 – 27] and among cortical areas [28 – 30].”
BackgroundSelf-citationA Spiking Neural Network Model of an Actor-Critic Learning Agent
Potjans, W., Morrison, A., & Diesmann, M. (2008) · Neural Computation · doi:10.1162/neco.2008.08-07-593 · Cited by 111
Same or Different? A Neural Circuit Mechanism of Similarity-Based Pattern Match Decision Making
Engel, T. A., & Wang, X. (2011) · Journal of Neuroscience · doi:10.1523/jneurosci.6150-10.2011 · Cited by 91 · cited 1×
“Other physiologically relevant phenomena such as bursting, adaptation, and oscillations are currently not captured by our framework, but can be incorporated in the future; adaptation, for example, can be included in phenomenological form appropriate to a rate model [81, 82].”
BackgroundSelf-citationPrefrontal Neural Correlates of Memory for Sequences
Averbeck, B. B., & Lee, D. (2007) · Journal of Neuroscience · doi:10.1523/jneurosci.4483-06.2007 · Cited by 85
Reinforcement Learning of Two-Joint Virtual Arm Reaching in a Computer Model of Sensorimotor Cortex
Neymotin, S. A., Chadderdon, G. L., Kerr, C. C., Francis, J. T., & Lytton, W. W. (2013) · Neural Computation · doi:10.1162/neco_a_00521 · Cited by 42
Linking the VPS35 and EIF4G1 Pathways in Parkinson’s Disease
Ross, O. A., Cook, C. N., & Petrucelli, L. (2015) · Neuron · doi:10.1016/j.neuron.2014.12.045 · Cited by 32
An objective function for STDP.Preprint
Bengio, Y., Mesnard, T., Fischer, A., Zhang, S., & Wu, Y. (2015) · arXiv (Cornell University) · Cited by 17
Analog Memories in a Balanced Rate-Based Network of E-I Neurons
Festa, D., Hennequin, G., & Lengyel, M. (2014) · Cited by 9
A Recurrent Network Model of Somatosensory Parametric Working Memory in the Prefrontal Cortex
Miller, P. C., Brody, C. D., Romo, R., & Wang, X. (2005) · Cerebral Cortex · doi:10.1093/cercor/bhi102 · Cited by 7
Self-citation
Research works printed but not linked: 2
- Bengio Y. 2015 Towards biologically plausible deep learning;Available from: http://arxiv.org/abs/1502.0415.
- DE Rumelhart, 1986, Parallel Distrib. Process, Vol. 1, 318
Cited by
Computation Through Neural Population Dynamics
Vyas, S., Golub, M. D., Sussillo, D., & Shenoy, K. V. (2020) · Annual Review of Neuroscience · doi:10.1146/annurev-neuro-092619-094115 · Cited by 671
“The authors developed a set of tools for reward-based training of RNNs, which match behavioral and electrophysiological recordings from an array of experimental paradigms and thus might serve as a powerful framework for studying value-based neural computations (Song et al. 2017).”
BackgroundFrontal Cortex and the Hierarchical Control of Behavior
Badre, D., & Nee, D. E. (2018) · Trends in Cognitive Sciences · doi:10.1016/j.tics.2017.11.005 · Cited by 611
“As noted above, many solutions surely exist to hierarchical control problems, including those not discussed here, such as recurrent networks [100].”
BackgroundTask representations in neural networks trained to perform many cognitive tasks
Yang, G. R., Joglekar, M. R., Song, H. F., Newsome, W. T., & Wang, X.-J. (2019) · Nature Neuroscience · doi:10.1038/s41593-018-0310-2 · Cited by 549
Self-citationArtificial Neural Networks for Neuroscientists: A Primer
Yang, G. R., & Wang, X.-J. (2020) · Neuron · doi:10.1016/j.neuron.2020.09.005 · Cited by 376
Self-citationWilliams, A. H., Kim, T. H., Wang, F., Vyas, S., Ryu, S. I., Shenoy, K. V., Schnitzer, M., Kolda, T. G., & Ganguli, S. (2018) · Neuron · doi:10.1016/j.neuron.2018.05.015 · Cited by 310
Metadata
- Authors
- Song, H. F., Yang, G. R., & Wang, X.-J.
- Affiliations
- New York University (US)
- New York University Shanghai (CN)
- Corresponding author
- Wang, Xiao-Jing
- Venue
- PLOS Computational Biology
- Publisher
- Public Library of Science (PLoS)
- Published
- 2016-02-29
- Volume
- 12
- Issue
- 2
- Pages
- e1004792
- ISSN
- 1553-734X (ISSN-L)
- 1553-7358
- PMC
- PMC4771709
- OpenAlex
- W2280725856
- Cited by
- 281
- References
- 91
- Citations by year
- 2026: 23 · 2025: 33 · 2024: 39 · 2023: 42 · 2022: 26 · 2021: 28 · 2020: 23 · 2019: 27 · 2018: 14 · 2017: 21 · 2016: 5
- FWCI
- 14.21
- Type
- journal article
- Language
- en
- Open access
- gold
- doi.org
- Licence
- Locations
- PLoS Computational Biology (publishedVersion, cc-by)
- PubMed Central (publishedVersion, cc-by)
- PubMed (publishedVersion)
- DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
- Figshare (submittedVersion, cc-by)
- Funders
- Office of Naval Research (N00014–13–1-0297, N00014)
- Topics
- Life Sciences › Neuroscience › Cognitive Neuroscience › Neural dynamics and brain function
- Physical Sciences › Computer Science › Artificial Intelligence › Neural Networks and Applications
- Life Sciences › Neuroscience › Cognitive Neuroscience › Memory and Neural Mechanisms
- Keywords
- Dale's principle, neural circuit model, cognitive tasks, perceptual decision making, multisensory integration, working memory, neural population dynamics, Theano, automatic differentiation, gradient-based learning
- MeSH terms
- Algorithms, Animals, Haplorhini, Cognition, Computer Simulation, Models, Neurological, Neurons, Computational Biology
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.