Optimal decision making in neural inhibition models.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Decision Making

Background:

• The drift diffusion model (DDM) is influential for understanding optimal decision making.
• Neural inhibition models, including the leaky competing accumulator (LCA) and feedforward inhibition (FFI) models, have been shown to mimic the DDM.
• Previous work suggested these neural models accomplish optimal decision making.

Purpose of the Study:

• To investigate the conditions under which neural inhibition models (LCA, FFI) accurately mimic the drift diffusion model (DDM).
• To examine the impact of handling negative activation values on the equivalence between neural models and the DDM.
• To propose a modification to neural models to ensure accurate DDM approximation.

Main Methods:

• Simulations were conducted to compare the performance of the DDM, truncated LCA, and truncated FFI.
• The study analyzed the effect of truncating negative activation values to zero in neural models.
• A baseline activity parameter was introduced into the LCA and FFI models.

Main Results:

• The equivalence between neural inhibition models and the DDM is contingent on the handling of negative activation values.
• Truncating negative activations to zero results in a loss of equivalence, with the DDM generally outperforming truncated LCA and FFI.
• Parameter estimates from truncated neural models no longer map simply onto DDM parameters.
• Introducing a baseline activity parameter allows the LCA to approximate the DDM and the FFI to be identical to the DDM.

Conclusions:

• The way neural inhibition models handle negative activations critically affects their ability to replicate the DDM.
• Truncation of negative activations leads to practical performance differences and complicates parameter mapping.
• Assuming a baseline activity in accumulators provides a neurophysiologically and mathematically sound method to achieve DDM equivalence for LCA and FFI models.