Press Esc to close

BODB


Hierarchy Browser (Show)

Model: Competitive Attractor Model (Wang)

Collator: Jimmy Bonaiuto
Created: September 05, 2016
Last modified by: Jimmy Bonaiuto
Last modified: November 17, 2016
Tags: Decision making, Neural network, leaky integrate-and-fire
Brief Description

The competitive attractor model uses competition between neural populations to choose between two alternate response options.

Public:  YES
Architecture
Inputs
  • noise (Poisson spikes) - Noisy background input
  • Ia (Spikes) - Task-related input signalling evidence for option A
  • Ib (Spikes) - Task-related input signalling evidence for option B
Outputs
  • decision (binary) - A or B
Diagrams (Show)
Figure 1: Model architecture
Model architecture

There are two pyramidal cell groups (A and B), each of which is selective to one of the two stimuli (mimicking motion to the right or left). Within each pyramidal neural group there is strong recurrent excitatory connections that can sustain persistent activity triggered by a transient preferred stimulus. The two pyramidal groups compete through feedback inhibition
from interneurons.

Figure 2: Model input
Model input

Top: the inputs are Poisson rates that vary in time and obey Gaussian distributions, with means muA and muB and with standard deviation sigma. The means muA and muB depend on the coherence level linearly (insert). Bottom: an example of stochastic inputs to neural groups A and B with mu0 = 40 and sigma=10 in Hz, coherence=6.4%. At every 50 ms, the two stimuli are independently resampled using their Gaussian distributions, so that the inputs vary stochastically in time. If sigma = 0, the two inputs would be constant in time.

 
Submodules (click to view and edit)
Narrative (Show)
Summaries of Experimental Data (SEDs) and Simulation Results (SSRs) (Show)
Related BOPs (Show)
Related Brain Regions (Show)
References (Show)
Discussion (Show)