Hosted by the Wellcome Neural Dynamics PhD Programme
Across the brain, neural activity recorded during behaviour simultaneously exhibits numerous sources of variability. For instance, even in sensory areas of the cortex, traditionally thought to be dedicated primarily to specific sensory modalities, actions or behavioural context can modulate activity. Additionally, sensory and behavioural variability is difficult to control, and perhaps even desirable to study ethologically relevant conditions. At the circuit level, not all variability in activity is equally relevant, some fluctuations may happen in input or output null-spaces, thus have no observable consequences. This complicates the interpretation of neural population activity. I will discuss an approach to extract neural dynamics related to external factors such as behaviour or stimulation. This approach is based on the hypothesis that neural circuits produce autonomous, low-dimensional dynamics, a direct result of their synaptic connectivity. I will demonstrate that it is possible to identify such dynamics with a machine learning technique called weak supervision. I will then show how such a model can be extended to reveal potential external inputs to a circuit. Finally, I will discuss some fundamental problems that arise in such models.
If you would like to join Matthias for lunch for lunch before the talk, please contact Jake Ahern.
If joining online: https://bristol-ac-uk.zoom.us/j/94138286231?pwd=MlRURE1SWjR6OTZCR1Fnak9QbGxhUT09, Meeting ID: 941 3828 6231 Passcode: 277162