Event-Based Computation for Touch Localization Based on Precise Spike Timing.

Area of Science:

• Neuroscience
• Artificial Intelligence
• Computational Biology

Background:

• Precise spike timing and temporal coding are crucial in biological nervous systems, but underutilized in conventional Artificial Neural Networks (ANNs).
• Spiking Neural Networks (SNNs) offer potential for temporal coding, yet practical applications outperforming rate-based ANNs remain an open challenge.
• Neuromorphic systems are well-suited for exploiting temporal coding due to their efficiency in processing spatio-temporal activity patterns.

Purpose of the Study:

• To propose and validate a neuromorphic model inspired by the sand scorpion to explore the benefits of temporal coding.
• To demonstrate the efficacy of temporal coding in an event-based sensory-processing task using unsupervised learning.
• To establish a benchmark task for evaluating event-based sensory processing models based on temporal coding.

Main Methods:

• Development of a neuromorphic model based on the sand scorpion's sensory processing.
• Validation of the model in a target localization task using relative spike timing from eight vibration sensors.
• Implementation of two unsupervised learning approaches for clustering spatio-temporal patterns within SNNs.

Main Results:

• Successful localization of a target by the proposed SNN models using only relative spike timing.
• Demonstration of task solvability through both analytical methods and numerical simulations of multiple SNN models.
• Validation of the model's capability to cluster spatio-temporal patterns in an unsupervised manner.

Conclusions:

• The proposed neuromorphic models effectively utilize precise spike timing for spatio-temporal pattern classification.
• The developed models offer an optimal approach for event-based sensory processing tasks relying on temporal coding.
• The target localization task serves as a valuable benchmark for assessing future event-based sensory processing models.