Motion detection based on recurrent network dynamics

Abstract

The detection of visual motion requires temporal delays to compare current with earlier visual input. Models of motion detection assume that these delays reside in separate classes of slow and fast thalamic cells, or slow and fast synaptic transmission. We used a data-driven modeling approach to generate a model that instead uses recurrent network dynamics with a single, fixed temporal integration window to implement the velocity computation. This model successfully reproduced the temporal response dynamics of a population of motion sensitive neurons in macaque middle temporal area (MT) and its constituent parts matched many of the properties found in the motion processing pathway (e.g., Gabor-like receptive fields (RFs), simple and complex cells, spatially asymmetric excitation and inhibition). Reverse correlation analysis revealed that a simplified network based on first and second order space-time correlations of the recurrent model behaved much like a feedforward motion energy (ME) model. The feedforward model, however, failed to capture the full speed tuning and direction selectivity properties based on higher than second order space-time correlations typically found in MT. These findings support the idea that recurrent network connectivity can create temporal delays to compute velocity. Moreover, the model explains why the motion detection system often behaves like a feedforward ME network, even though the anatomical evidence strongly suggests that this network should be dominated by recurrent feedback.

Testing the assumptions underlying fMRI adaptation using intracortical recordings in area MT

“…Second, conceptually area MT appears to be operate at a stage of visual processing that is more involved with the extraction of features than the combination of features into invariant representations of objects (Fujita, 2002). MT’s emphasis on extraction is consistent with computational models (Adelson & Bergen, 1985; Joukes et al, 2014) and experimental data that reveal competitive interactions (Gaudio & Huang, 2012; Krekelberg & Albright, 2005; Krekelberg & van Wezel, 2013; Xiao et al, 2014), the weighing of evidence and counter-evidence (Duijnhouwer & Krekelberg, 2015), and segmentation of figure and ground (X. Huang, Albright, & Stoner, 2007; X.…”

“…it is not direction selective) could become direction selective after adaptation of its inputs (Tolias, Keliris, Smirnakis, & Logothetis). These are just some examples of how modulating the output of a subset of neurons in a recurrently connected network can have complex consequences and emphasizes that tuning is not a property of a single neuron but emerges from an interconnected network of neurons (Joukes, Hartmann, & Krekelberg, 2014; Richert, Albright, & Krekelberg, 2013). …”

Adaptation without Plasticity

“…Standard motion models typically assume delay lines or postulate the existence of classes of neurons that are intrinsically slow or fast. In the recurrent network, however, these properties emerge from the network dynamics (Joukes et al, 2014). Other examples of the computational flexibility of recurrent networks include the selective amplification of noisy signals (Hahnloser et al, 2002), and the state-dependent processing of inputs (Rutishauser and Douglas, 2009).…”

Evidence and Counterevidence in Motion Perception

“…For instance, neurons in area MT are sensitive to the higher-order correlations created by multiple successive motion steps in the same direction (Mikami et al 1986), and both fruit flies and humans reliably extract additional motion information from scenes containing three-point diverging and converging spatiotemporal correlations that are invisible to the standard motion models and the Bours-Lankheet model (Hu and Victor 2010;Fitzgerald et al 2011;Clark et al 2014). Recent work from our laboratory suggests that networks of recurrently connected neurons are well suited to extract such higher-order statistical regularities dynamically (Richert et al 2013;Joukes et al 2014). …”