Learning by Delay Modifications

In order to account for real time processing by the visual system, it has been proposed that single spike arrival times encode analog information, which can be processed faster than firing rates by a lateral inhibitory network. We show that learning can occur in such a scheme by introducing variable delays, resulting in both fast and adaptive processing. The A-current, able to control delays, and known to be involved in learning, is proposed as a biological substrate. The model offers a complementary mechanism for modeling receptive field plasticity.