THE COMPUTATIONAL MODELING OF WORKING MEMORY
DECAY IN THE VISUOSPATIAL DELAYED-RESPONSE EXPERIMENTS
Miha PELKO, Grega REPOVŠ
1 Jožef Stefan international Postgraduate School, Ljubljana, Slovenia
2 Department of Psychology, Faculty of Arts, University of Ljubljana, Slovenia
INTRODUCTION Target Delay Response The subjects maintain fixation on a
central point while a peripheral target
Fix X X
is flashed for 200ms at a certain angle.
Sustained firing of neurons observed in neuropsychological Following a delay of 2-5s, the fixation
experiments in monkeys is thought to be the neuronal correlate of the point is extinguished and the subjects
make a memory-guided saccade to
representations in working memory. Those representations are the original location of the target.
temporary, where the proposed mechanisms for the degradation of
working memory are temporal decay and interference [1,2]. We While only behavioral information
(angle of the saccade) is measured in
specifically focus on the computational modeling of visuospatial the experiments involving humans ,
delayed-response experiments, where the experiments on humans also singe-cell measurements are
taken in the experiments with monkeys
 and monkeys , performing a memory-guided saccade task show 200ms
, more precisely the neural activity of
stable accuracy, but decreased precision of with time. An appropriate Fix
the cells corresponding to a specific
computational model should: angle.
? to encode persistent stable activity with decreased
precision even in the absence of stimuli.
? system of memory degradation coherent with the findings
in the single-cell studies.
To simulate the neuronal activity of the prefrontal cortex during the delayed-response task we use one
of the most frequently found model in the literature from Compte et al. . This is conductance based
integrate-and-fire model, including GABA (inhibitory), AMPA and NMDA (both excitatory) mediated
The neural network in the model is composed of the populations of inhibitory and excitatory neurons
(EN), and external noisy stimulus representing the input from other brain areas. All the neurons are
EN are encoding the information of the target angle (the memory). Each neuron represents a specific
target angle. The connections among EN are formed in a way that the neurons encoding similar
angles are better connected (higher synaptic connectivity), then the ones encoding different angles.
The chart in the upper right shows such an exemplary profile of connectivity for a neuron encoding
The simulations were preformed using the Brian Simulator .
The tested model successfully produces a stable persistent
? model we were unable to fully simulate the decrease of
the neural activity in accordance with the single-cell measurements.
More thorough systematic approach in the search within the mode
connectivity parameters is required.
The sequences of two of the performed simulations for different sets of connectivity parameters are
shown above (top and bottom).
The simulated neurons (N=2560) are constantly receiving random external stimuli. At t=200ms we induce
a current to the neurons encoding the primary direction (PD = 180°) and the neighboring neurons for the
period of 200ms, representing visual stimulus. This is followed by the delay period, when the neurons are
again only receiving the background stimulation and the input from each other (fully interconnected
network), which enables the persistent encoding.
The raster plots on the left represent the activity (spike frequency) of the excitatory neuronal population  Portrat, S., Barrouillet, P., & Camos, V. (2008). Time-related decay or interference-based forgetting in
encoding the target angle over time. The recorded spikes are averaged over the time interval t=240ms working memory? Journal of Experimental Psychology. Learning, Memory, and Cognition, 34(6), 1561-
and the angle of 6°. The left figures present the activity of the neurons encoding the primary direction (PD - 1564. doi: 10.1037/a0013356.
red lines) and the neurons encoding the two other angles (155°- green lines, 135° - blue lines ). The dotted
lines roughly represent the schematic expected activity from the single-cell measurements.  Oberauer, K., & Lewandowsky, S. (2008). Forgetting in Immediate Serial Recall: Decay, Temporal
Distinctiveness, or Interference? Psychological Review, 115(3), 576, 544.
For the first simulation (top) we set a strong excitatory interconnectivity. The network encodes the target
angle and holds this stable encoding over the whole delay period. The lack of any degradation of neural  Ploner, C. J., Gaymard, B., Rivaud, S., Agid, Y., & Pierrot-Deseilligny, C. (1998). Temporal limits of spatial
working memory in humans. The European Journal of Neuroscience, 10(2), 794-797.
activity can be observed on the right figure.
 White, R.L. & Snyder, L.H. (2005). Dynamics of Memory-related Spatial Tuning in the Frontal Eye Field.
The second simulation employs a more conservative parameter setting, generally resulting in lower
activity. However, as we can see on the right figure, the activity for all observed neurons only drops  Compte, A., Brunel, N., Goldman-Rakic, P. S., & Wang, X. (2000). Synaptic Mechanisms and Network
immediately after we shut down the cue current, after which it remains constant on average. Dynamics Underlying Spatial Working Memory in a Cortical Network Model. Cereb. Cortex, 10(9), 910-923.