SAM: A Unified Self-Adaptive Multicompartmental Spiking Neuron Model for Learning With Working Memory

Area of Science:

• Computational Neuroscience
• Artificial Intelligence
• Neuromorphic Engineering

Background:

• Working memory is crucial for biological cognition and advanced artificial intelligence.
• Endowing simple neuron models with working memory and understanding its neuronal basis remains challenging.

Purpose of the Study:

• To present a novel self-adaptive multicompartment spiking neuron model (SAM) for spike-based learning with integrated working memory.
• To explore the biological principles and parameters influencing SAM's dynamics and working memory capabilities.
• To evaluate SAM's performance on diverse cognitive tasks, highlighting its efficiency and robustness.

Main Methods:

• Developed a novel self-adaptive multicompartment spiking neuron model (SAM) incorporating sparse coding, dendritic non-linearity, intrinsic self-adaptive dynamics, and spike-driven learning.
• Constructed spiking networks using the SAM model to perform tasks such as MNIST supervised learning, noisy spike pattern classification, and meta-learning.
• Analyzed the impact of SAM model variations on its working memory functionality.

Main Results:

• The SAM model successfully integrated spike-driven learning and working memory within a single neuron framework.
• SAM-based networks demonstrated energy efficiency and robustness across various challenging tasks, including sequential learning and meta-learning.
• Experimental results provide insights into the neuronal mechanisms underlying working memory.

Conclusions:

• The SAM model represents a significant advancement in creating biologically plausible artificial neurons with working memory capabilities.
• Its competitive performance and energy efficiency position it as a valuable component for developing advanced neuromorphic computing systems.
• This work offers potential insights into the biological underpinnings of working memory in neural systems.