The student will study the relationship between the topology and behaviour of networks in the context of adaptation through learning. He/she will focus on random random recurrent network models (e.g., liquid state machines, random population networks) that are augmented with learning. In SORN, unsupervised learning (intrinsic plasticity and STDP) was used to adapt random networks of binary threshold neurons to specific input patterns. Later, this was augmented with reward modulation to achieve supervised learning. Although the authors study activity patterns in the resulting networks, the focus of these works is mainly on the performance of these networks for, e.g., performing tasks.
The student will study how the local topology and local neural adaptation mechanisms in biological networks (e.g., threshold adaptation, intrinsic plasticity) affect the global network behaviour and the efficiency with which the network can be optimised (learn) to perform the desired behaviour. For this purpose, he/she will develop models to quantify and compare the behaviour and information processing in networks with local unsupervised (IP and STDP) and supervised (reward modulation) learning. These will then be used to study the sensitivity of network evolution during learning to parameters of the learning rules. In view of developing future biologically inspired computing algorithms and systems, he/she will also evaluate how such mechanisms can be efficiently implemented in analog or digital hardware.
More information: https://www.ugent.be/en/work/vacancies/scientific/phd-student-lo8js
Maass, W. et al. (2002). Neural Comput. 14, 2531–2560.
Lazar, A. et al. (2009). Frontiers in Comput. Neuroscience. DOI: 10.3389/neuro.10.023.2009
Witali, A. et al. (2015). Frontiers in Comput. Neuroscience. DOI: 10.3389/fncom.2015.00036
1. Model-based quantification of the evolution of network behaviour and information processing during learning.
2. Model how learning rules and learning rule parameters affect the resulting network behaviour, information processing and supervised learning efficiency.
3. Evaluation of efficiency with which learning rules can be implemented in hardware.
1. IMBB, FORTH, Panayiota Poirazi, M15-18: learning and adaptation in biological networks.
2. Easics, M25-M27, Ramses Valvekens: hardware implementation for embedded learning.
Enrolment in Doctoral degree(s): You will be enrolled at Ghent University PhD programme