On diffusion processes with variable drift rates as models for decision making during learning.

Area of Science:

• Cognitive Neuroscience
• Computational Neuroscience
• Decision Science

Background:

• Evidence accumulation models, like the diffusion model, are crucial for understanding decision-making.
• Understanding how learning affects the parameters of these models is key to explaining behavioral improvements.

Purpose of the Study:

• To model evidence accumulation in a visual discrimination task using Ornstein-Uhlenbeck and diffusion processes with variable drift rates.
• To characterize how model parameters influence performance accuracy and how they change with learning.

Main Methods:

• Derivation of power-law and exponential drift-rate models.
• Fitting these models to psychophysical data from monkeys during task learning.
• Analysis of parameter changes reflecting learning effects on the decision process.

Main Results:

• The overall drift rate, representing the signal-to-noise ratio of sensory evidence, increased steadily with training.
• Secondary parameters related to the timing of drift during stimulus viewing did not show consistent trends.
• Simple diffusion model versions effectively captured behavioral changes across training.

Conclusions:

• Learning in this visual task primarily enhances the efficiency of evidence processing (drift rate).
• The diffusion model provides a quantitative framework for understanding learning-related changes in decision-making.
• Model parameters offer insights into the neural mechanisms underlying decision improvement.