Designing spiking neural networks for robust and reconfigurable computation

Area of Science:

• Computational neuroscience
• Artificial intelligence
• Complex systems

Background:

• Spiking neural networks (SNNs) offer resilient analog computing.
• Symmetrically connected inhibitory SNNs can perform computations robustly, even with neuron loss.
• Adapting network parameters for disruption resilience remains a challenge.

Purpose of the Study:

• To develop an analytical approach for deriving network parameters in SNNs.
• To investigate the dynamics of k-winners-takes-all (k-WTA) computations in SNNs.
• To enable the design of disruption-resilient SNN computing systems.

Main Methods:

• Analysis of k-winners-takes-all (k-WTA) computations in SNNs.
• Characterization of different dynamical regimes within the network.
• Derivation of analytical expressions for transitions between k-winner states.

Main Results:

• Identification of distinct dynamical regimes in k-WTA computations.
• Analytical expressions provided for parameter transitions based on input and network properties.
• Demonstration of how to adapt parameters for robust k-WTA functionality.

Conclusions:

• The study offers analytical insights into the dynamics of k-WTA computations.
• Provides a method for designing SNNs with disruption-resilient dynamics.
• Enhances understanding of parameter adaptation for robust SNN computing.