“…Simultaneous imaging of the surrounding network has allowed us to show that, despite this exquisite behavioural sensitivity, the dominant network response matches the activation of targeted neurons with local suppression, flattening the network input-output function to maintain the level of network activation within the spontaneous range. These results support the sparse coding hypothesis (Barlow, 1972;Barth & Poulet, 2012;Kanerva, 1993;Olshausen & Field, 1996), demonstrate that the local network operates in an inhibition-stabilized regime (Denève & Machens, 2016;Murphy & Miller, 2009;Ozeki, Finn, Schaffer, Miller, & Ferster, 2009;Pehlevan & Sompolinsky, 2014;Sanzeni et al, 2020;Tsodyks et al, 1997;van Vreeswijk & Sompolinsky, 1996;Wolf et al, 2014) and suggest a high storage capacity for recurrent networks (Hopfield, 1982;Lefort et al, 2009;Peron et al, 2020). This combination of features likely maximizes perceptual sensitivity while minimizing erroneous detection of background activity, thus avoiding hallucinations (Carbon, 2014;Cassidy et al, 2018;Corlett et al, 2019;Friston, 2005), runaway excitation (Rose & Blakemore, 1974;Treiman, 2001;Ziburkus, Cressman, Barreto, & Schiff, 2006) and reducing cortical energy requirements (Schölvinck et al, 2008).…”