Inspired by Francis Collins on guitar and all the wonderful speakers at Stand Up For Science DC today. Keep it up!

(8/11) Thus, we suggest that cortex can learn contextual, predictive motor control from subcortical “demonstration” of innate behavior. This doesn’t seem too crazy since imitation learning has been uniquely successful for efficiently teaching robots how to control their bodies.

(5/11) Those jump movements are powerful and make detailed recordings and manipulations prohibitive, so we designed a rig that enabled a head-fixed mouse to instead push a platform down using hindlimb extension. They get a cue that predicts cold air, and can quickly learn to avoid it:

(3/11) The body is complex, so learning to control it quickly in a new situation is a vexing problem. Fortunately, cerebral cortex is at the top of a hierarchy of behavioral control and can “re-use” subcortical components, for example if they can already control a jump.

(2/11) We were inspired by how quickly both young and old mice could learn to jump to escape a cold air stream: