Implementing Signature Neural Networks with Spiking Neurons

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Bio-inspired Computing

Background:

• Spiking Neural Networks (SNNs) use precise spike timing for information coding, offering a more biologically realistic alternative to traditional Artificial Neural Networks (ANNs).
• Recent neuroscience discoveries highlight novel computational principles in living neural systems that could inspire advancements in ANNs.
• Existing SNN research has not yet incorporated mechanisms like neural signatures, local contextualization, and multicoding strategies.

Purpose of the Study:

• To adapt and apply the core concepts of the Signature Neural Network paradigm to Spiking Neural Networks.
• To provide a proof-of-concept for the applicability of neural signatures, local information contextualization, and multicoding in SNNs.
• To explore how these bio-inspired mechanisms enhance self-organization, information processing, and memory capabilities in SNNs.

Main Methods:

• Implementation of neural signatures for unit identification within an SNN.
• Integration of local information contextualization for data processing.
• Application of multicoding strategies for information propagation in the SNN.
• Computer simulations to analyze network self-organization and emergent properties.

Main Results:

• The adapted SNN model demonstrated complex self-organizing properties.
• Multiple simultaneous encoding schemes enabled the generation of coexisting spatio-temporal patterns.
• Competition among patterns emerged without inhibitory connections, modulated by network and intra-unit parameters.
• Parameters also influenced the network's memory capabilities, with independent dynamical modes across informational dimensions.

Conclusions:

• Plasticity mechanisms and multicoding strategies can confer additional computational properties to SNNs.
• These bio-inspired approaches have the potential to enhance SNN capacity and performance for diverse real-world applications.
• The study validates the applicability of Signature Neural Network concepts in advancing SNNs.