cfaed Seminar Series presents speech on “Effects of unstructured input on self-organized neural network”
Presentation of a Bachelor Thesis
cfaed Seminar Series presents speech on “Effects of unstructured input on self-organized neural network” by Matthias Loidolt, TU Dresden. The presentation will be held on February 24, 2020, 13:30 o'clock at the TU Dresden, Barkhausen-Bau, Room BAR IV75 (Helmholtzstr. 18, 01069 Dresden). Everybody is welcome!
For more information about the topic, please consider the abstract below:
“Effects of unstructured input on self-organized neural network”
An unresolved issue of cortical development is whether spontaneous activity prepares the cortex for
sensory input. This thesis focuses on the development of sequence memory, a basic ability of cortex
underlying sensory perception. The cortex is modeled as a recurrent neural network equipped with
homeostatic and spike-timing-dependent plasticity (STDP). This model has been shown to perform
visual sequence replay when learning from structured input. Here, I show that even learning from
unstructured input increases general sequence memory performance. Moreover, such ‘pre’-learning
from unstructured input speeds up the subsequent learning of specific sequences. The key structural
substrate is the emergence of few strong and directed synapses due to STDP and synaptic
competition. These construct self-amplifying preferential paths of activity, which can subsequently
be harnessed to quickly memorize new input sequences. This thesis therefore proposes that
endogeneous sequences arising from spontaneous activity form the backbone of sequence memory.