One of the key insights of systems neuroscience is that neurons support decision making as experts: individual neurons are sensitive to specific, sometimes impressively complex, features of the environment. These neurons can then signal to the next stages of processing the challenges we are actually facing with. However, later research has revealed that a neuron is not the ‘specialist’ we thought it would be, instead these neurons are eagerly multi-tasking. This phenomenon, known as mixed selectivity, might sound beneficial as neurons are not limited to do a single job. Mixed selectivity, however, poses a daunting challenge both for processing layers reading the output of these neurons, and for us to be able to decode the signals the information neurons are actually carrying because we cannot be certain which environmental signal the given neuron was signalling. In a collaboration with the Golshani lab (UCLA) we demonstrated that despite mixed selectivity sensory information can be reliably decoded along with other task relevant variables. We demonstrated the even in the primary visual cortex a representation is in place that can accommodate many relevant features without the danger of interference, such that different relevant variables (including content of visual stimulus) are encoded in orthogonal subspaces. Here, while inidividual neurons show selectivity to a multitude of variables, we show that there are specific combinations of individual activities which unambiguously identify sensory and cognitive variables as well.

This can be though of a cube floating in the space with different kind of information represented on the different faces. For each kind of inforamtion one can find an angle, looking from where only one specific type of information is available without the distraction of other pieces of information.