A prediction of the model is that stimulation during the transition period (but not before) would affect RTs. We showed this by microstimulating FEF, and found that pre-planned eye movements could be forced to be faster than normal with stimulation (purple below). (7/8)

We built a model of the system, to make sense of our empirical observations. In the model, RT variability is the result of the amplification of noise during motor preparation period (blue) by the geometry of the manifold. (6/8)

A small proportion of the variance associated with RT was explained by the initial conditions during movement preparation. However, a large proportion was explained by the trajectory length (but not the speed) of the population activity through the transition subspace. (5/8)

We found that movement preparation signals are bumped out of their attractor state by go-cue signals (~60 to 85ms post-go-cue). The network then moves through a “transition subspace” towards a “movement-initiation subspace” that emerged ~35ms before the movement. (4/8)

Building on work from multiple labs (a non-exhaustive list includes work from @MarkChurchland, @svoboda314, @mjaz_jazlab, @LabSchall, @SmithLabNeuro, and their colleagues), we used the dynamical systems approach to analyze population data from the frontal eye field. (3/8)

Also panel F looks like the "after" of this restoration. The figure does a great job at conveying the main results though. Beautiful paper.

Oh wow, a trillion notifications. I don't know if this was brought up already, but isn't Lundqvist... Miller 2016/2018 good support for a specific WM model that requires STP? In particular, the observation that within gamma bursts there is more info than outside:

(7/7) Finally, by separately analyzing the population properties of different regions, we found that posterior regions play a particularly important role in working memory, in sharp contrast with the lack of task-related responses in the anterior DLPFC.

(6/7) We found that this global functional gradient is better described by an overall trend that emerges from the organized arrangement of distinct anatomical regions, as opposed to a smooth gradient that spans multiple anatomical regions.

(5/7) We assessed whether this global gradient was better explained by a smooth change of functional properties across multiple anatomical regions, or by an abrupt change occurring at the border of adjacent regions.