A continuous-time neural model for sequential action.

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Modeling

Background:

• Sequential action control involves continuous processes influenced by perception and internal goals.
• Existing models are either hierarchical with hand-built representations or heterarchical lacking goal-orientation.
• The theory of event coding (TEC) proposes shared representations for actions and perceptions.

Purpose of the Study:

• To present a biologically motivated, neurocomputational model for sequential action control.
• To implement the theory of event coding (TEC) within a neural architecture.
• To enable goal-directed learning and dynamic predictions in hierarchically structured tasks.

Main Methods:

• Developed a model within the Leabra neural architecture, incorporating unsupervised and goal-directed learning.
• Embedded the neurocomputational model within the theoretical framework of the theory of event coding (TEC).
• Utilized continuous-time inputs and generated non-stationary outputs for dynamic predictions.

Main Results:

• The model successfully implements TEC for sequential action control in tasks like coffee-making.
• It demonstrates the capability for both unsupervised and goal-directed learning.
• The model generates short-timescale dynamic predictions, unlike traditional static models.

Conclusions:

• The proposed model offers a biologically plausible approach to understanding and controlling sequential actions.
• It bridges the gap between perception and action through shared representations as proposed by TEC.
• This framework supports dynamic, goal-oriented behavior in complex, hierarchical tasks.