Recurrent circuit based neural population codes for stimulus representation and inference

A large part of the synaptic input received by cortical neurons comes from local cortico-cortical connectivity. Despite their abundance, the role of local recurrence in cortical function is unclear, and in simple coding schemes it is often the case that a circuit with no recurrent connections performs optimally. We consider a recurrent excitatory-inhibitory circuit model of a cortical hypercolumn which performs sampling-based Bayesian inference to infer latent hierarchical stimulus features. We show that local recurrent connections can store an internal model of the correlations between stimulus features that are present in the external world. When the resulting recurrent input is combined with feedforward input it produces a population code from which the posterior over the stimulus features can be linearly read out. Internal Poisson spiking variability provides the proper fluctuations for the population to sample stimulus features, yet the resultant population variability is aligned along the stimulus feature direction, producing what are termed differential correlations. Importantly, the amplitude of these internally generated differential correlations is determined by the associative prior in the model stored in the recurrent connections, thus providing experimentally testable predictions for how population connectivity and response variability are connected to the structure of latent external stimuli.

The Devil Is in the Detail: Brain Dynamics in Preparation for a Global–Local Task

When analyzing visual scenes, it is sometimes important to determine the relevant “grain” size. Attention control mechanisms may help direct our processing to the intended grain size. Here we used the event-related optical signal, a method possessing high temporal and spatial resolution, to examine the involvement of brain structures within the dorsal attention network (DAN) and the visual processing network (VPN) in preparation for the appropriate level of analysis. Behavioral data indicate that the small features of a hierarchical stimulus (local condition) are more difficult to process than the large features (global condition). Consistent with this finding, cues predicting a local trial were associated with greater DAN activation. This activity was bilateral but more pronounced in the left hemisphere, where it showed a frontal-to-parietal progression over time. Furthermore, the amount of DAN activation, especially in the left hemisphere and in parietal regions, was predictive of subsequent performance. Although local cues elicited left-lateralized DAN activity, no preponderantly right activity was observed for global cues; however, the data indicated an interaction between level of analysis (local vs. global) and hemisphere in VPN. They further showed that local processing involves structures in the ventral VPN, whereas global processing involves structures in the dorsal VPN. These results indicate that in our study preparation for analyzing different size features is an asymmetric process, in which greater preparation is required to focus on small rather than large features, perhaps because of their lesser salience. This preparation involves the same DAN used for other attention control operations.

The Effects of Unilateral Pulvinar Damage in Humans on Reflexive Orienting and Filtering of Irrelevant Information

The effects of damage to the pulvinar nucleus of the thalamus in humans on reflexive orienting and selective attention were investigated. In a spatial orienting task three patients with unilateral pulvinar damage determined the location of a visual target that followed a cue that was not informative as to the targets location. Contralesional targets were responded to more slowly than ipsilesional targets. Also, at long cue target intervals patients responses to contralesional targets that appeared at previously cued locations were slower than to non-cued locations indicating that pulvinar damage does not affect inhibition of return. In the selective attention task two of the patients identified a target that appeared at one level of a global-local hierarchical stimulus while ignoring a distractor present at the other level. The distractor indicated either the same response as the target or a different response. Response times to targets in both visual fields were similar as were interference effects from the ignored distractors. These data indicate that engaging attention contralesionally is not impaired in discrimination tasks and that filtering of irrelevant information was not impaired contralesionally.