Parametrization of learning tasks for the analysis of switch-dependent memory consolidation
Promotor(s) : Drion, Guillaume
Date of defense : 26-Jun-2023/27-Jun-2023 • Permalink :
|Parametrization of learning tasks for the analysis of switch-dependent memory consolidation
|Date of defense :
|Committee's member(s) :
|Engineering, computing & technology > Multidisciplinary, general & others
|Université de Liège, Liège, Belgique
|Master en ingénieur civil biomédical, à finalité spécialisée
|Master thesis of the Faculté des Sciences appliquées
[en] [en] When we learn something new, our brain initially forms a fragile memory. Then, this memory needs to be consolidated so that it can be stored more permanently. Memory consolidation describes the process by which our memories strengthen to become durable in our brain. But what influences this process?
One hypothesis of consolidation is the impact of activity switches on the reinforcement of neuronal connections. This property of neurons encoding information is known as synaptic plasticity. Experimental evidence in cognitive neuroscience has shed light on the mechanisms by which plasticity is triggered. By employing modeling techniques that replicate these experimental findings, we take another approach to understanding the processes underlying memory.
Experimental studies have provided compelling evidence for the correlation between the transition of neuronal firing patterns in the brain and the formation and consolidation of memory.
To study the impact of these different activities, it is necessary to define a learning task. However, the literature on plasticity modeling is extensive. To gain insight into how to design a learning task, a literature review is conducted. The implementation of these tools is then combined with the choices of free parameters related to the tasks. The mechanisms of potentiation and depression based on the frequencies of neuronal input signals are also explored.
By applying this parametrization, limiting phenomenons of models have been highlighted, namely the spike transmission and its influence on network connectivity. This refers to the "erroneous" increase in synaptic plasticity for non-activated presynaptic neurons. From these limitations, implementations of solutions have been explored. Two of them made it possible to overcome this problem. The first one was the definition of a maximum synaptic capacity, and the second one was a simplified approach to a homeostasis mechanism for the excitability of postsynaptic neurons.
Finally, once parameterization was done and limitations were overcome, learning tasks on neuronal circuits were carried out. First, the network was presented with a grid pattern without overlapped pixels and showed a consolidation consistent with memory shaping. Then, patterns with overlapped pixels were shown. Depending on the simulation parameters chosen, two results were observed: total or partial consolidation of overlapped pixels.
Such sensitive parameterization of plasticity rules can be seen on the one hand as an inherent fragility of those models.
On the other hand, the differences between the results obtained by changing the parameters could reflect the different situations occurring in the brain under the action of external parameters such as mood.
The question of the right learning method remains open and merits further investigation. A noted perspective might be to combine the strengths of several research fields.
Cite this master thesis
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.