G.F.C. (Guillaume) Etter, Dr

My research focuses on learning and memory, and uses humans, animals, and artificial neural networks as models of cognition.

1) Biological learning rules

For the past 30 years, artificial neural networks have been trained using error backpropagation, which has some limitations and is not biologically plausible. This prompts a need to identify how biological neural networks leverage inhibitory interneurons, neuromodulators, as well as top-down circuits to compute errors and update synapses to improve behavior. Additionaly, what is the exact role of neural activity during sleep? How can biological and artificial systems aquire new knowledge in absence of supervision or reward signals?

2) Brain-inspired continual learning

Learning and memory require a tight balance between acquiring new information over time through experience while retaining previous knowledge. This challenge is commonly referred to as the stability-plasticity trade-off. Animals have the remarkable ability to leverage previous knowledge to quickly solve new tasks with little trial-and-error. On the other hand, current artificial systems based on neural networks are unable to learn from continuous streams of data and tend to forget previous knowledge when learning from new data distributions. This is an important issues as it makes current state-of-the-art AI models one-use only, thus posing energy and sustainability concerns. By combining neuroscience experimentation and AI models, this project aims to elucidate the mechanisms of continual learning in the brain and translate those findings to AI models.

3) Novel diagnoses and treatments for neurodegenerative disorders

Alzheimer's disease is characterized by cognitive decline and progressive neurodegenerescence. Strikingly, several hallmarks can be used to predict cognitive decline long before cell death can be observed. These hallmarks include electrophysiological changes related to pathological inhibitory neurons that can be picked up by EEG, a low-cost, non-invasive recording method. Using state-of-the-art AI models, this project aims to develop early diagnoses for Alzheimer's disease. Additionally, using minimally invasive optogenetic stimulation approaches, this project aims to develop novel stimulation-based approaches to restore memory function in Alzheimer's disease conditions.