Spiking neural models for decision-making tasks with learning.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Neural Networks

Background:

• Drift Diffusion Models (DDMs) and Poisson counter models are standard for decision-making, but lack learning mechanisms.
• Existing models are limited to tasks with prior category knowledge, hindering biological integration.
• Bridging cognitive and biological models requires incorporating learning and realistic neural dynamics.

Purpose of the Study:

• To propose a biologically plausible Spiking Neural Network (SNN) model for decision-making with an integrated learning mechanism.
• To model neural activity using a multivariate Hawkes process, linking it to cognitive processes.
• To bridge the gap between cognitive and biological models of decision-making.

Main Methods:

• Established a mathematical coupling between the Drift Diffusion Model (DDM) and the Poisson counter model.
• Demonstrated that DDMs can be approximated by spiking Poisson neurons.
• Derived a specific DDM with correlated noise from a Hawkes network of spiking neurons with a local learning rule.
• Designed an online categorization task to empirically validate model predictions.

Main Results:

• Showed that DDMs and Poisson counter models yield similar categorizations and reaction times.
• Confirmed that the DDM can be approximated by spiking Poisson neurons.
• Successfully derived a DDM with correlated noise from a Hawkes network model governed by a local learning rule.
• Empirical evaluation of the model predictions using an online categorization task.

Conclusions:

• The proposed Spiking Neural Network (SNN) model offers a biologically plausible framework for decision-making.
• This work integrates neural mechanisms into cognitive models, enhancing understanding of neural activity and behavior.
• The findings represent a significant step toward unifying cognitive and neural theories of decision-making.