A Deep Learning Approach to Analyzing Continuous-Time Cognitive Processes.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Artificial Intelligence

Background:

• Cognitive processes are dynamic and influenced by numerous variables, especially in natural settings.
• Traditional statistical models often oversimplify these complex dynamics.
• Deep learning excels at simulating cognitive processes but lacks interpretability for scientific analysis.

Purpose of the Study:

• To develop an interpretable deep learning model for analyzing complex cognitive dynamics.
• To bridge the gap between deep learning's simulation power and scientific analytical needs.
• To relax implausible assumptions common in standard cognitive data analysis.

Main Methods:

• Developed a nonlinear regression model using artificial neural networks.
• Parameterized probability distributions by convolving predictor histories over time.
• Inferred the shape and temporal extent of effects directly from time-series data.

Main Results:

• Demonstrated significant improvements on behavioral and neuroimaging data in language processing.
• The model successfully relaxed assumptions of linearity, stationarity, and homoscedasticity.
• Enabled discovery of novel patterns and control for confounds in analyses.

Conclusions:

• Deep learning can be effectively used for analyzing, not just simulating, complex cognitive processes.
• The proposed model offers interpretable function approximation for time-series cognitive data.
• This approach opens new avenues for studying cognitive (neuro)science phenomena.