Robust computation with rhythmic spike patterns.

Area of Science:

• Computational Neuroscience
• Neuromorphic Engineering
• Theoretical Neuroscience

Background:

• Spike-timing coding offers faster, more energy-efficient information processing than rate coding.
• Spike-timing codes are often brittle, limiting their application in neuroscience and computing.
• Attractor neural networks provide a framework for robust memory and computation.

Purpose of the Study:

• To propose a novel attractor neural network model for robust spiking neural networks.
• To demonstrate a phase-to-timing mapping for computational properties in spiking neural networks.
• To explore implications for neuroscience and neuromorphic computing.

Main Methods:

• Developed threshold phasor associative memory (TPAM) networks storing complex phasor patterns.
• Derived the energy function governing TPAM's fixed-point attractor dynamics.
• Constructed two spiking neural network models (reductionist and biologically plausible) to approximate TPAM computations.

Main Results:

• TPAM networks exhibit high memory capacity with sparse phasor patterns.
• Fixed points in TPAM correspond to stable, perturbation-robust periodic spiking activity.
• Spiking neural networks successfully approximated TPAM's complex algebraic computations.

Conclusions:

• Established a link between rhythmic firing patterns and complex attractor dynamics.
• Proposed a framework for robust computation in spiking neural networks.
• Findings have implications for interpreting neural spike patterns and designing neuromorphic devices.