Recurrent Neural Circuits Overcome Partial Inactivation by Compensation and Re-learning.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Artificial Intelligence

Background:

• Advances in artificial neural manipulation spur research into brain circuit causality.
• Interpreting experimental results is challenging due to neural circuit complexity.
• New theoretical frameworks are needed for causal effect reasoning in neuroscience.

Purpose of the Study:

• To develop a theoretical framework for understanding causal effects in neural circuits.
• To explain the magnitude of behavioral effects from perturbations using dynamical system structure.
• To identify strategies for improving the interpretability of inactivation experiments.

Main Methods:

• Utilized recurrent neural networks trained on perceptual decision-making tasks.
• Analyzed the dynamical system structure underlying network solutions.
• Modeled the effects of perturbations on network behavior.

Main Results:

• Network dynamical structure precisely accounts for behavioral effects of perturbations.
• The framework explains how complex circuits compensate and adapt to perturbations.
• Identified sensitive behavioral features and strategies to improve inactivation experiment interpretability.

Conclusions:

• Understanding dynamical systems in neural networks offers a precise account of causal effects.
• The proposed framework clarifies neural circuit compensation and adaptation mechanisms.
• This approach enhances the interpretability of neural inactivation studies in cognitive neuroscience.