GCAL Development Demo

Jean-Luc R. Stevens, Judith S. Law, Jan Antolik, and James A. Bednar

Self-organization

(A version of this page with smaller pictures is also available. )

This animation shows:

(top left) Ferret V1 orientation maps, recorded by Chapman et al. (1996), over approximately a 10-day period using chronic optical imaging. As selectivity emerges, orientation preferences change very little.

(top right) GCAL model orientation maps from Stevens et al. (2013), over approximately 8000 training iterations at 25% contrast, showing that the development of orientation preference in this model is stable over time, as in the ferret data.

(bottom right) Afferent weights to approximately every third neuron in the L model (trained at 25% contrast), colored by orientation preference. Even for this optimal contrast value, development of orientation preference in this simpler model is highly unstable, unlike in the ferret data.

(bottom left) Same plot as for bottom right, but for the GCAL model. Adding contrast-gain control and homeostatic adaptation of excitability to the L model allows stable development of preferences for orientation for each neuron.

The animation repeats 3 times to so that viewers can inspect each segment of the plot, and at the very end the afferent weights for the GCAL model neurons are shown in a larger size. Overall, these plots demonstrate how neurons and maps can develop stable preferences for different orientations, as observed in the animal data, if the model includes contrast-gain control and homeostasis.


References