Diffusive information accumulation by minimal recurrent neural models of decision making

Area of Science:

• Computational Neuroscience
• Cognitive Psychology
• Decision Making Models

Background:

• Psychological models of decision making often assume evidence accumulation via diffusion to a response criterion.
• These diffusion models successfully explain reaction time (RT) distributions and choice probabilities in various tasks.
• A key theoretical challenge is identifying the neural basis for the diffusive evidence accumulation process.

Purpose of the Study:

• To investigate a potential neural implementation of diffusive evidence accumulation in decision making.
• To analyze a simple recurrent neural architecture for its capacity to generate diffusion processes.
• To evaluate if this neural architecture can reproduce empirical findings in decision-making tasks.

Main Methods:

• Analysis of a simple recurrent neural network architecture designed for decision making.
• Mathematical modeling of the evidence accumulation process within this architecture.
• Characterization of the resulting diffusion process as a time-inhomogeneous Ornstein-Uhlenbeck velocity process.

Main Results:

• The analyzed recurrent architecture naturally implements a diffusive evidence accumulation process.
• This process is mathematically described as a time-inhomogeneous Ornstein-Uhlenbeck velocity process.
• The model's predictions for RT distributions and choice probabilities closely match behavioral data.

Conclusions:

• Persistent activity in reverberation loops, as suggested by Wang (2001, 2002), can neurally realize diffusive evidence accumulation.
• The proposed recurrent neural architecture provides a viable mechanism for implementing established decision-making models.
• This framework offers a strong link between neural dynamics and psychological theories of choice and reaction time.