Topic Summary. Recent advances in high density neural recording technology
render it possible to record from many hundreds of neurons simultaneously
from the brains of animals as they perform behavioral and cognitive tasks.
The unit of observation is no longer the action potentials from one well
characterized neuron, but the action potentials from N neurons. There are
many ways to characterize such activity. A popular approach relies on
weighted sums of the spike counts, or rates, from all neurons, where the
weights establish a coding direction in the N-dimensional neuronal state
space (SS). The SS characterization of neural activity leans heavily on
Machine Learning (decoding) and linear algebra (e.g., geometry of low
dimensional manifolds) - methods that are also popular in human
neuroimaging (e.g., voxel-based decoding). In addition to a healthy dose
of intuition, the speakers will illustrate ways in which SS characterizations
elucidate and/or obfuscate neural mechanisms in their own work.