Learning recurrent dynamics in spiking networks

Kim, Christopher M.; Chow, C. C.

doi:10.1101/297424

Cited by 14 publications

(30 citation statements)

References 64 publications

(177 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead, some of them were relying on control theory to train a chaotic reservoir of spiking neurons [32][33][34] . Others used the FORCE algorithm 35,36 or variants of it 35,[37][38][39] . However, the FORCE algorithm was not argued to be biologically realistic, as the plasticity rule for each synaptic weight requires knowledge of the current values of all other synaptic weights.…”

Section: Discussionmentioning

confidence: 99%

A solution to the learning dilemma for recurrent networks of spiking neurons

et al. 2020

View full text Add to dashboard Cite

Recurrently connected networks of spiking neurons underlie the astounding information processing capabilities of the brain. Yet in spite of extensive research, how they can learn through synaptic plasticity to carry out complex network computations remains unclear. We argue that two pieces of this puzzle were provided by experimental data from neuroscience. A mathematical result tells us how these pieces need to be combined to enable biologically plausible online network learning through gradient descent, in particular deep reinforcement learning. This learning method-called e-prop-approaches the performance of backpropagation through time (BPTT), the best-known method for training recurrent neural networks in machine learning. In addition, it suggests a method for powerful on-chip learning in energy-efficient spike-based hardware for artificial intelligence.

show abstract

Section: Discussionmentioning

confidence: 99%

A solution to the learning dilemma for recurrent networks of spiking neurons

et al. 2020

View full text Add to dashboard Cite

show abstract

“…To circumvent this issue, some networks have held fixed inhibitory efficacies [ 64 , 99 – 101 ]. Others have attempted to obey Dale’s principle by suddenly freezing synaptic connections if the synaptic update reversed the sign of excitatory or inhibitory influences [ 102 ]. In subsequent trainings, synapses attempting to change their signs were excluded and therefore prevented from exhibiting activity-dependent changes.…”

Section: Discussionmentioning

confidence: 99%

Embodied working memory during ongoing input streams

2021

View full text Add to dashboard Cite

Sensory stimuli endow animals with the ability to generate an internal representation. This representation can be maintained for a certain duration in the absence of previously elicited inputs. The reliance on an internal representation rather than purely on the basis of external stimuli is a hallmark feature of higher-order functions such as working memory. Patterns of neural activity produced in response to sensory inputs can continue long after the disappearance of previous inputs. Experimental and theoretical studies have largely invested in understanding how animals faithfully maintain sensory representations during ongoing reverberations of neural activity. However, these studies have focused on preassigned protocols of stimulus presentation, leaving out by default the possibility of exploring how the content of working memory interacts with ongoing input streams. Here, we study working memory using a network of spiking neurons with dynamic synapses subject to short-term and long-term synaptic plasticity. The formal model is embodied in a physical robot as a companion approach under which neuronal activity is directly linked to motor output. The artificial agent is used as a methodological tool for studying the formation of working memory capacity. To this end, we devise a keyboard listening framework to delineate the context under which working memory content is (1) refined, (2) overwritten or (3) resisted by ongoing new input streams. Ultimately, this study takes a neurorobotic perspective to resurface the long-standing implication of working memory in flexible cognition.

show abstract

“…One such method is based on the first-order reduced and controlled error (FORCE) algorithm previously developed for rate RNNs [6]. The FORCE-based methods are capable of training spiking networks, but training all the parameters including recurrent connections could become computationally inefficient [21][22][23]. Lastly, recent studies successfully converted rate-based networks trained with a gradient-descent method to spiking networks for both convolutional and recurrent neural networks [24,25].…”

Section: Introductionmentioning

confidence: 99%

Learning the synaptic and intrinsic membrane dynamics underlying working memory in spiking neural network models

Kim

Sejnowski

2020

Preprint

View full text Add to dashboard Cite

Recurrent neural network (RNN) model trained to perform cognitive tasks is a useful computational tool for understanding how cortical circuits execute complex computations. However, these models are often composed of units that interact with one another using continuous signals and overlook parameters intrinsic to spiking neurons. Here, we developed a method to directly train not only synaptic-related variables but also membrane-related parameters of a spiking RNN model.Training our model on a wide range of cognitive tasks resulted in diverse yet task-specific synaptic and membrane parameters. We also show that fast membrane time constants and slow synaptic decay dynamics naturally emerge from our model when it is trained on tasks associated with working memory (WM). Further dissecting the optimized parameters revealed that fast membrane properties and slow synaptic dynamics are important for encoding stimuli and WM maintenance, respectively. This approach offers a unique window into how connectivity patterns and intrinsic neuronal properties contribute to complex dynamics in neural populations. Results 40Here, we provide an overview of the method that we developed to directly train spiking recurrent 41 neural network (RNN) models (for more details see Methods). Throughout the study, we considered 42 recurrent network models composed of leaky integrate-and-fire (LIF) units whose membrane voltage 43 dynamics were governed by: 44where τ m,i is the membrane time constant of unit i, v i (t) is the membrane voltage of unit i at time 45 t, v rest,i is the resting potential of unit i, and R i is the input resistance of unit i. I i (t) represents 46 the current input to unit i at time t, which is given by: 47 −1 ( Fig. 2A; see Methods). The two input stimuli for the DIS task were sinusoidal waves with 129 different frequencies, modeled after the task used by Romo et al. [32] where monkeys were trained 130 to discriminate two vibratory stimuli. If the first stimulus had a higher (lower) frequency, our RNN 131 model was trained to produce a positive (negative) output signal ( Fig. 2B; see Methods). 132It took longer to train our model on these two tasks compared to the delayed integration task 133 (7103.95 ± 3738.65 trials for the DMS task and 6985.47 ± 2112.34 trials for the DIS task). The 134 distributions of the tuned parameters from the two WM tasks were similar to the distributions 135 obtained from the delayed integration task (Fig. 2C and D). More importantly, we again observed 136 significantly faster membrane voltage dynamics and slower synaptic decay from the inhibitory 137 6

show abstract

Learning recurrent dynamics in spiking networks

Cited by 14 publications

References 64 publications

A solution to the learning dilemma for recurrent networks of spiking neurons

A solution to the learning dilemma for recurrent networks of spiking neurons

Embodied working memory during ongoing input streams

Learning the synaptic and intrinsic membrane dynamics underlying working memory in spiking neural network models

Contact Info

Product

Resources

About