P086 Synergistic short-term synaptic plasticity mechanisms for working memory
Florian Fiebig*1, Nikolaos Chrysanthisdis1, Anders Lansner1,2, Pawel Herman1,3
1KTH Royal Institute of Technology, Dept of Computational Science and Technology, Stockholm, Sweden
2Stockholm University, Department of Mathematics, Stockholm, Sweden
3Digital Futures, KTH Royal Institute of Technology
*Email: fiebig@kth.se
Introduction
Working memory (WM) is essential for almost every cognitive task. The neural and synaptic mechanisms supporting the rapid encoding and maintenance of memories in diverse tasks are the subject of ongoing debate. The traditional view of WM as stationary persistent firing of selective neuronal populations has given room to newer ideas for mechanisms that support a more dynamic maintenance of multiple items, that may also tolerate activity disruption. Computational WM models based on different biologically plausible synaptic and neural plasticity mechanisms have been proposed but not combined systematically. Monolithic models (WM function explained by one particular mechanism) are theoretically appealing but also narrow explanations.
Methods
In this study we evaluate the interactions between three commonly used classes of plasticity: Intrinsic excitability (postsynaptic, increasing the excitability of spiking neurons), synaptic facilitation/augmentation (presynaptic, potentiating outgoing synapses of spiking neurons) and Hebbian plasticity (pre-post-synaptic, potentiating recurrent synapses driven by correlations), see Fig.1. Combinations of these mechanisms are systematically tested in a spiking neural network model on a broad suite of tasks or functional motifs deemed principally important for WM operation, such as one-shot encoding, free and cued recall, delay maintenance and updating. In our evaluation we focus on operational task performance and biological plausibility.
Results
We show that previously proposed short-term plasticity mechanisms may not necessarily be competing explanations, but instead yield interesting functional interactions on a wide set of WM tasks and enhance the biological plausibility of spiking neural network models. Our results indicate that a composite model, combining several commonly proposed plasticity mechanisms for WM function, is superior to more reductionist variants. Importantly, we attribute the observable differences to the principle nature of specific types of plasticity. For example, we find a previously undescribed synergistic function of Hebbian plasticity that supports the rapid updating of multi-item WM sets through rapidly learned inhibition.
Discussion
Our study suggests that commonly used forms of plasticity proposed for the buffering of WM information besides persistent activity are eminently compatible, and yield synergies that improve function and biological plausibility in a modular spiking neural network model. Combinations enable a more holistic model of WM responsive to broader task demands than what can be achieved with more reductionist models. Conversely, the targeted ablation of specific plasticity components reveals that different mechanisms are differentially important to specific aspects of WM function, advancing the search for more capable, robust and flexible models accounting for new experimental evidence of bursty and activity-silent multi-item maintenance.
Figure 1. Fig.1-Plasticity Combinations. The Augmentation plasticity model is implemented using the well-known Tsodyks-Makram mechanism [1]. The Bayesian Confidence Propagation Neural Network (BCPNN) learning rule implements intrinsic plasticity, as well as Hebbian plasticity [2]. These 3 components can be simulated separately or together, yielding 7 scenarios to simulate and study.
Acknowledgements
We would like to thank the Swedish Research Council (VR) grants: 2018-05360 and 2016-05871, Digital Futures and Swedish e-science Research Center (SeRC) for their support.
References
Tsodyks, M., Pawelzik, K., & Markram, H. (1998). Neural Networks with Dynamic Synapses.Neural Computation,10(4), 821–835.
Tully, P. J., Hennig, M. H., & Lansner, A. (2014). Synaptic and nonsynaptic plasticity approximating probabilistic inference.Frontiers in Synaptic Neuroscience,6, 8.
