The incremental learning algorithm for compartmental spiking neuron model

Abstract

Abstract Within the framework of the neuromorphic approach, a compartmental spiking neuron model was developed. The compartmental spiking neuron model was used to solve the classification problem using a small training set. However, despite the biological inspiration of the model, the used compartmental spiking neuron model was unable to learn new instances online. The structural learning algorithm used limited the model to use only in offline scenarios, while there are a large number of tasks where the ability to adapt to new data coming in during model operation and the ability to work with data distributions that change over time are necessary. Based on this, the task of online restructuring of the model is relevant. In this paper, we propose a new algorithm for training a compartmental spiking neuron model, which allows the model to be used in incremental learning scenarios.