Malleable temporal integration of positional information for moving objects.

One of the primary functions of visual processing is to generate a spatial mapping of our immediate vicinity, in order to facilitate interaction. As yet it is unclear how this is achieved, but the process likely involves an accrual of information over time--a temporal integration of positional information (Eagleman & Sejnowski, 2000; Krekelberg et al., 2000). Temporal integration is a common computational process evident in diverse settings, such as electrical engineering (Bryson & Ho, 1975) and neural coding (Rao, Eagleman & Sejnowski, 2001; Usher & McClelland, 2001). In the later context it is sometimes assumed that integration dynamics are immobile, and consequently that they can be diagnostic of a sensory system (Arnold & Lipp, 2011; Krauskopf & Mollon, 1971; Snowden & Braddick, 1991). Other data suggest that integration times can be flexible, varying in concert with the properties of a stimulus (Bair & Movshon, 2004) or environment (Ossmy et al., 2013). Our data provide behavioral support for malleable integration times. We examine a motion-induced illusion of perceived position linked to temporal integration, and use prolonged exposure to motion of different speeds (sensory adaptation) to modulate the dynamics of neural activity. Results show that perceived position is governed by a weighted average of positional estimates from multiple channels with distinct, fixed integration times. Postadaptation channel contributions are reweighted, resulting in coding that is optimized to the dynamics of the prevailing environment.