Explicit melioration by a neural diffusion model.

Area of Science:

• Behavioral neuroscience
• Cognitive psychology
• Computational neuroscience

Background:

• Animals adjust response rates between two reward sources to increase overall reinforcement.
• Herrnstein's matching law describes the equilibrium where relative response rates match relative reinforcement rates.
• Melioration was proposed as the dynamic process underlying this matching behavior but lacked detailed mechanisms.

Purpose of the Study:

• To specify the dynamical process of melioration by integrating decision-making and operant conditioning theories.
• To extend the drift diffusion model (DDM) to predict both choice proportions and inter-response times.
• To explore neural network implementations and their implications for brain organization.

Main Methods:

• Implemented melioration within an adaptive drift diffusion model (DDM).
• Varied DDM parameters (drift and threshold) to investigate their relationship with reward rates.
• Analyzed predictions for choice probability and inter-response times from DDM variants.
• Considered neural network models of the proposed mechanism.

Main Results:

• An adaptive DDM with specific parameter settings dynamically tracks exact matching behavior.
• DDMs with fixed thresholds and reward-rate-dependent drift approximate matching but yield different predictions.
• The models offer insights into how reward rate estimation via leaky integration could be implemented neurally.

Conclusions:

• Melioration and matching may arise from synaptic processes estimating reward rates through leaky integration.
• This mechanism links input and output stages in a two-stage stimulus-response system.
• The findings provide a computational framework for understanding reward-based choice behavior and its neural basis.