Slow manifolds within network dynamics encode working memory efficiently and robustly

Working memory is a cognitive function involving the storage and manipulation of latent information over brief intervals of time, thus making it crucial for context-dependent computation. Here, we use a top-down modeling approach to examine network-level mechanisms of working memory, an enigmatic issue and central topic of study in neuroscience and machine intelligence. We train thousands of recurrent neural networks on a working memory task and then perform dynamical systems analysis on the ensuing optimized networks, wherein we find that four distinct dynamical mechanisms can emerge. In particular, we show the prevalence of a mechanism in which memories are encoded along slow stable manifolds in the network state space, leading to a phasic neuronal activation profile during memory periods. In contrast to mechanisms in which memories are directly encoded at stable attractors, these networks naturally forget stimuli over time. Despite this seeming functional disadvantage, they are more efficient in terms of how they leverage their attractor landscape and

[1]  Christos Constantinidis,et al.  Stable population coding for working memory coexists with heterogeneous neural dynamics in prefrontal cortex , 2016, Proceedings of the National Academy of Sciences.

[2]  Omri Barak,et al.  One Step Back, Two Steps Forward: Interference and Learning in Recurrent Neural Networks , 2018, Neural Computation.

[3]  M. R. Riley,et al.  Role of Prefrontal Persistent Activity in Working Memory , 2016, Front. Syst. Neurosci..

[4]  Srdjan Ostojic,et al.  Coding with transient trajectories in recurrent neural networks , 2018, PLoS Comput. Biol..

[5]  Xiao-Jing Wang.  Synaptic reverberation underlying mnemonic persistent activity , 2001, Trends in Neurosciences.

[6]  Surya Ganguli,et al.  Universality and individuality in neural dynamics across large populations of recurrent networks , 2019, NeurIPS.

[7]  Christopher D. Harvey,et al.  Recurrent Network Models of Sequence Generation and Memory , 2016, Neuron.

[8]  N. Cowan.  What are the differences between long-term, short-term, and working memory? , 2008, Progress in brain research.

[9]  Wenwen Bai,et al.  Dynamic trajectory of multiple single-unit activity during working memory task in rats , 2015, Front. Comput. Neurosci..

[10]  Wulfram Gerstner,et al.  Stability of working memory in continuous attractor networks under the control of short-term plasticity , 2018, bioRxiv.

[11]  M. Tsodyks,et al.  Working models of working memory , 2014, Current Opinion in Neurobiology.

[12]  Christian Tetzlaff,et al.  Working Memory Requires a Combination of Transient and Attractor-Dominated Dynamics to Process Unreliably Timed Inputs , 2017, Scientific Reports.

[13]  Francesca Mastrogiuseppe,et al.  Linking Connectivity, Dynamics, and Computations in Low-Rank Recurrent Neural Networks , 2017, Neuron.

[14]  Daniel B. Rubin,et al.  The Stabilized Supralinear Network: A Unifying Circuit Motif Underlying Multi-Input Integration in Sensory Cortex , 2015, Neuron.

[15]  E. Rolls,et al.  Attention and working memory: a dynamical model of neuronal activity in the prefrontal cortex , 2003, The European journal of neuroscience.

[16]  L. F. Abbott,et al.  Generating Coherent Patterns of Activity from Chaotic Neural Networks , 2009, Neuron.

[17]  L. Abbott,et al.  Eigenvalue spectra of random matrices for neural networks. , 2006, Physical review letters.

[18]  P. Goldman-Rakic,et al.  Synaptic mechanisms and network dynamics underlying spatial working memory in a cortical network model. , 2000, Cerebral cortex.

[19]  J J Hopfield,et al.  Neural networks and physical systems with emergent collective computational abilities. , 1982, Proceedings of the National Academy of Sciences of the United States of America.

[20]  S. Funahashi,et al.  Stable and Dynamic Coding for Working Memory in Primate Prefrontal Cortex , 2017, The Journal of Neuroscience.

[21]  Erin L. Rich,et al.  Stable and dynamic representations of value in the prefrontal cortex , 2019, bioRxiv.

[22]  Peter Ford Dominey,et al.  Reservoir Computing Properties of Neural Dynamics in Prefrontal Cortex , 2016, PLoS Comput. Biol..

[23]  K. Sakai.  Task set and prefrontal cortex. , 2008, Annual review of neuroscience.

[24]  Xiao-Jing Wang,et al.  Task representations in neural networks trained to perform many cognitive tasks , 2019, Nature Neuroscience.

[25]  Joel Z. Leibo,et al.  Prefrontal cortex as a meta-reinforcement learning system , 2018, bioRxiv.

[26]  Christopher D. Harvey,et al.  Choice-specific sequences in parietal cortex during a virtual-navigation decision task , 2012, Nature.

[27]  R. Romo,et al.  Neuronal correlates of parametric working memory in the prefrontal cortex , 1999, Nature.

[28]  C. Curtis,et al.  Persistent activity in the prefrontal cortex during working memory , 2003, Trends in Cognitive Sciences.

[29]  Danna Zhou,et al.  d. , 1840, Microbial pathogenesis.

[30]  Guangyu R. Yang,et al.  Training Excitatory-Inhibitory Recurrent Neural Networks for Cognitive Tasks: A Simple and Flexible Framework , 2016, PLoS Comput. Biol..

[31]  Matthew T. Kaufman,et al.  A neural network that finds a naturalistic solution for the production of muscle activity , 2015, Nature Neuroscience.

[32]  Wolfgang Maass,et al.  Emergence of complex computational structures from chaotic neural networks through reward-modulated Hebbian learning. , 2014, Cerebral cortex.

[33]  Razvan Pascanu,et al.  A neurodynamical model for working memory , 2011, Neural Networks.

[34]  Surya Ganguli,et al.  Reverse engineering recurrent networks for sentiment classification reveals line attractor dynamics , 2019, NeurIPS.

[35]  Sommers,et al.  Chaos in random neural networks. , 1988, Physical review letters.

[36]  David J. Freedman,et al.  Computing by Robust Transience: How the Fronto-Parietal Network Performs Sequential, Category-Based Decisions , 2017, Neuron.

[37]  Steven W Kennerley,et al.  Reconciling persistent and dynamic hypotheses of working memory coding in prefrontal cortex , 2017, Nature Communications.

[38]  Masud Husain,et al.  Neural mechanisms of attending to items in working memory , 2019, Neuroscience & Biobehavioral Reviews.

[39]  Guangyu Robert Yang,et al.  Artificial Neural Networks for Neuroscientists: A Primer , 2020, Neuron.

[40]  O. Barak,et al.  Dynamics of random recurrent networks with correlated low-rank structure , 2019, 1909.04358.

[41]  J. Fuster,et al.  Delayed-matching and delayed-response deficit from cooling dorsolateral prefrontal cortex in monkeys. , 1976, Journal of comparative and physiological psychology.

[42]  L. Abbott,et al.  From fixed points to chaos: Three models of delayed discrimination , 2013, Progress in Neurobiology.

[43]  S. Cavanagh,et al.  Reconciling persistent and dynamic hypotheses of working memory coding in prefrontal cortex , 2017, bioRxiv.

[44]  P. Goldman-Rakic,et al.  Mnemonic coding of visual space in the monkey's dorsolateral prefrontal cortex. , 1989, Journal of neurophysiology.

[45]  Danielle S. Bassett,et al.  Multimodal network dynamics underpinning working memory , 2020, Nature Communications.

[46]  Dmitri B. Chklovskii,et al.  Neuronal Circuits Underlying Persistent Representations Despite Time Varying Activity , 2012, Current Biology.

[47]  M. Jung,et al.  Dynamically changing neuronal activity supporting working memory for predictable and unpredictable durations , 2019, Scientific Reports.

[48]  Wei Ji Ma,et al.  A diverse range of factors affect the nature of neural representations underlying short-term memory , 2018, Nature Neuroscience.

[49]  J. Zylberberg,et al.  Mechanisms of Persistent Activity in Cortical Circuits: Possible Neural Substrates for Working Memory. , 2017, Annual review of neuroscience.

[50]  Earl K. Miller,et al.  Working Memory 2.0 , 2018, Neuron.

[51]  Rishidev Chaudhuri,et al.  Computational principles of memory , 2016, Nature Neuroscience.