Control of criticality and computation in spiking neuromorphic networks with plasticity

Abstract

The critical state is assumed to be optimal for any computation in recurrent neural networks, because criticality maximizes a number of abstract computational properties. We challenge this assumption by evaluating the performance of a spiking recurrent neural network on a set of tasks of varying complexity at - and away from critical network dynamics. To that end, we developed a spiking network with synaptic plasticity on a neuromorphic chip. We show that the distance to criticality can be easily adapted by changing the input strength, and then demonstrate a clear relation between criticality, task-performance and information-theoretic fingerprint. Whereas the information-theoretic measures all show that network capacity is maximal at criticality, this is not the case for performance on specific tasks: Only the complex, memory-intensive tasks profit from criticality, whereas the simple tasks suffer from it. Thereby, we challenge the general assumption that criticality would be beneficial for any task, and provide instead an understanding of how the collective network state should be tuned to task requirement to achieve optimal performance.

References in corpus (19)

• Power-law distributions in empirical data
• Measuring Information Transfer
• Powerlaw: a Python package for analysis of heavy-tailed distributions
• Optimal Dynamical Range of Excitable Networks at Criticality
• A scalable multi-core architecture with heterogeneous memory structures for Dynamic Neuromorphic Asynchronous Processors (DYNAPs)
• Dynamical synapses causing self-organized criticality in neural networks
• Colloquium: Criticality and dynamical scaling in living systems
• JIDT: An information-theoretic toolkit for studying the dynamics of complex systems
• Self-Organized Branching Processes: A Mean-Field Theory for Avalanches
• Partial Information Decomposition as a Unified Approach to the Specification of Neural Goal Functions
• Adaptive self-organization in a realistic neural network model
• Bits from Biology for Computational Intelligence
• Demonstrating Advantages of Neuromorphic Computation: A Pilot Study
• Inferring collective dynamical states from widely unobserved systems
• An Accelerated LIF Neuronal Network Array for a Large Scale Mixed-Signal Neuromorphic Architecture
• Shared Information -- New Insights and Problems in Decomposing Information in Complex Systems
• Dynamic Adaptive Computation: Tuning network states to task requirements
• BROJA-2PID: A robust estimator for bivariate partial information decomposition
• Tailored ensembles of neural networks optimize sensitivity to stimulus statistics

Cited by in corpus (13)

• Segregation, integration and balance of large-scale resting brain networks configure different cognitive abilities
• Self-organization toward criticality by synaptic plasticity
• A Survey on Reservoir Computing and its Interdisciplinary Applications Beyond Traditional Machine Learning
• Modular architecture facilitates noise-driven control of synchrony in neuronal networks
• Hopf Bifurcation in Mean Field Explains Critical Avalanches in Excitation-Inhibition Balanced Neuronal Networks: A Mechanism for Multiscale Variability
• Tackling the subsampling problem to infer collective properties from limited data
• MR. Estimator, a toolbox to determine intrinsic timescales from subsampled spiking activity
• BioSFQ circuit family for neuromorphic computing: Bridging digital and analog domains of superconductor technologies
• Homeostatic Criticality in Neuronal Networks
• When to be critical? Performance and evolvability in different regimes of neural Ising agents
• Neuronal correlations shape the scaling behavior of memory capacity and nonlinear computational capability of reservoir recurrent neural networks
• Optimal reinforcement learning near the edge of synchronization transition
• 2022 Roadmap on Neuromorphic Computing and Engineering