Learning and Memory Strategies in Evolving Environments

from the immune system to neural networks

by Oskar H. Schnaack
Doctoral thesis
Date of Examination:2021-12-10
Date of issue:2022-01-17
Advisor:Dr. Armita, Nourmohammad
Referee:Dr. Armita, Nourmohammad
Referee:Prof. Dr. Stefan Klumpp
Persistent Address: http://dx.doi.org/10.53846/goediss-9015

 

 

Files in this item

Name:thesis.pdf
Size:6.98Mb
Format:PDF

The following license files are associated with this item:

Abstract

English

Storing memory of molecular encounters is vital for an effective response to recurring external stimuli. Interestingly, memory strategies vary among different biological processes. These strategies range from networks that process input signals and retrieve an associative memory to specialized receptors that bind only to related stimuli. The adaptive immune system uses such a specialized strategy and can provide specific responses against many pathogens. During its response, the immune system retains some cells as memory to act quicker when reinfections with the same or evolved pathogens occur. However, differentiation of memory cells remains one of the least understood cell fate decisions in immunology. The ability of immune memory to recognize evolved pathogens makes it an ideal starting point to study learning and memory strategies for evolving environments—a topic with applications far beyond immunology. In this thesis, I present three projects that study different aspects of memory strategies for evolving stimuli. Indeed, we find that specialized memory strategies can follow the evolution of stimuli and reliably recover memory of previous encounters. In contrast, fully connected networks, such as Hopfield networks, fail to reliably recover the memory of evolving stimuli. Thus, pathogen evolution might be the reason that the immune system produces specialized memories. We further find that specialized memory receptors should trade off their maximal binding for cross-reactivity to bind to evolved targets. To produce such receptors, the differentiation into memory cells in the immune system should be highly regulated. Finally, we study update strategies of memory repertoires using an energy-based model. We find that repertoires should have a moderate risk tolerance to fluctuations in performance to adapt to the evolution of targets. Nevertheless, these systems can be very efficient in distinguishing between evolved versions of stored targets and novel random stimuli.
Keywords: adaptive immune system; non-equilibrium decision-making; Hopfield network; memory strategies
 
Statistik