Modeling feature-based attention as an active top-down inference process

Vision is a crucial sensor. It provides a very rich collection of information about our environment. The difficulty in vision arises, since this information is not obvious in the image, it has to be constructed. Wheres earlier approaches have favored a bottom-up approach, which maps the image onto an internal representation of the world, more recent approaches search for alternatives and develop frameworks which make use of top-down connections. In these approaches vision is inherently a constructive process which makes use of a priory information. Following this line of research, a model of primate object perception is presented and used to simulate an object detection task in natural scenes. The model predicts that early responses in extrastriate visual areas are modulated by the visual goal.     

Visual attention: bottom-up versus top-down

Visual attention is attracted by salient stimuli that 'pop out' from their surroundings. Attention can also be voluntarily directed to objects of current importance to the observer. What happens in the brain when these two processes interact?     

Time course of attention reveals different mechanisms for spatial and feature-based attention in area V4

Attention can facilitate visual processing, emphasizing specific locations and highlighting stimuli containing specific features. To dissociate the mechanisms of spatial and feature-based attention, we compared the time course of visually evoked responses under different attention conditions. We recorded from single neurons in area V4 during a delayed match-to-sample task that controlled both spatial and feature-based attention. Neuronal responses increased when spatial attention was directed toward the receptive field and were modulated by the identity of the target of feature-based attention. Modulation by spatial attention was weaker during the early portion of the visual response and stronger during the later portion of the response. In contrast, modulation by feature-based attention was relatively constant throughout the response. It appears that stimulus onset transients disrupt spatial attention, but not feature attention. We conclude that spatial attention reflects a combination of stimulus-driven and goal-driven processes, while feature-based attention is purely goal driven.     

Action understanding and active inference

We have suggested that the mirror-neuron system might be usefully understood as implementing Bayes-optimal perception of actions emitted by oneself or others. To substantiate this claim, we present neuronal simulations that show the same representations can prescribe motor behavior and encode motor intentions during action?Cobservation. These simulations are based on the free-energy formulation of active inference, which is formally related to predictive coding. In this scheme, (generalised) states of the world are represented as trajectories. When these states include motor trajectories they implicitly entail intentions (future motor states). Optimizing the representation of these intentions enables predictive coding in a prospective sense. Crucially, the same generative models used to make predictions can be deployed to predict the actions of self or others by simply changing the bias or precision (i.e. attention) afforded to proprioceptive signals. We illustrate these points using simulations of handwriting to illustrate neuronally plausible generation and recognition of itinerant (wandering) motor trajectories. We then use the same simulations to produce synthetic electrophysiological responses to violations of intentional expectations. Our results affirm that a Bayes-optimal approach provides a principled framework, which accommodates current thinking about the mirror-neuron system. Furthermore, it endorses the general formulation of action as active inference.     

Searching for the neural mechanisms of feature-based attention in the primate brain

In this issue, two studies, one by Zhou and Desimone and another by Cohen and Maunsell, provide new insights into the mechanisms of feature-based attention (FBA). The former demonstrates a new role of the frontal eye fields in the origins of FBA and the latter shows that FBA is coordinated across both hemispheres.     

Frontal-to-parietal top-down causal streams along the dorsal attention network exclusively mediate voluntary orienting of attention

Previous effective connectivity analyses of functional magnetic resonance imaging (fMRI) have revealed dynamic causal streams along the dorsal attention network (DAN) during voluntary attentional control in the human brain. During resting state, however, fMRI has shown that the DAN is also intrinsically configured by functional connectivity, even in the absence of explicit task demands, and that may conflict with effective connectivity studies. To resolve this contradiction, we performed an effective connectivity analysis based on partial Granger causality (pGC) on event-related fMRI data during Posner's cueing paradigm while optimizing experimental and imaging parameters for pGC analysis. Analysis by pGC can factor out exogenous or latent influences due to unmeasured variables. Typical regions along the DAN with greater activation during orienting than withholding of attention were selected as regions of interest (ROIs). pGC analysis on fMRI data from the ROIs showed that frontal-to-parietal top-down causal streams along the DAN appeared during (voluntary) orienting, but not during other, less-attentive and/or resting-like conditions. These results demonstrate that these causal streams along the DAN exclusively mediate voluntary covert orienting. These findings suggest that neural representations of attention in frontal regions are at the top of the hierarchy of the DAN for embodying voluntary attentional control.     

The scope and limits of top-down attention in unconscious visual processing

Attentional selection plays a critical role in conscious perception. When attention is diverted, even salient stimuli fail to reach visual awareness. Attention can be voluntarily directed to a spatial location or a visual feature for facilitating the processing of information relevant to current goals. In everyday situations, attention and awareness are tightly coupled. This has led some to suggest that attention and awareness might be based on a common neural foundation, whereas others argue that they are mediated by distinct mechanisms. A body of evidence shows that visual stimuli can be processed at multiple stages of the visual-processing streams without evoking visual awareness. To illuminate the relationship between visual attention and conscious perception, we investigated whether top-down attention can target and modulate the neural representations of unconsciously processed visual stimuli. Our experiments show that spatial attention can target only consciously perceived stimuli, whereas feature-based attention can modulate the processing of invisible stimuli. The attentional modulation of unconscious signals implies that attention and awareness can be dissociated, challenging a simplistic view of the boundary between conscious and unconscious visual processing.     

Dopamine, affordance and active inference

The role of dopamine in behaviour and decision-making is often cast in terms of reinforcement learning and optimal decision theory. Here, we present an alternative view that frames the physiology of dopamine in terms of Bayes-optimal behaviour. In this account, dopamine controls the precision or salience of (external or internal) cues that engender action. In other words, dopamine balances bottom-up sensory information and top-down prior beliefs when making hierarchical inferences (predictions) about cues that have affordance. In this paper, we focus on the consequences of changing tonic levels of dopamine firing using simulations of cued sequential movements. Crucially, the predictions driving movements are based upon a hierarchical generative model that infers the context in which movements are made. This means that we can confuse agents by changing the context (order) in which cues are presented. These simulations provide a (Bayes-optimal) model of contextual uncertainty and set switching that can be quantified in terms of behavioural and electrophysiological responses. Furthermore, one can simulate dopaminergic lesions (by changing the precision of prediction errors) to produce pathological behaviours that are reminiscent of those seen in neurological disorders such as Parkinson's disease. We use these simulations to demonstrate how a single functional role for dopamine at the synaptic level can manifest in different ways at the behavioural level.     

A source for category-induced global effects of feature-based attention in human prefrontal cortex

1. Download : Download high-res image (181KB)
2. Download : Download full-size image

     

Hierarchical partitioning as an interpretative tool in multivariate inference

Abstract Much of biogeography, conservation and evolutionary biology, and ecology involves very large spatial and temporal extents. Direct manipulation to test hypotheses is usually almost impossible at appropriate scales so that multivariate modelling and especially regression are used to draw causal inferences about which ‘independent’ variables influence the distribution and abundances of species. Such inferences clearly are crucial for the successful management of biological resources and for conserving threatened species. A succession of regression approaches has arisen, many of which yield inconsistent implications. The main problem has been the quest for one (the ‘best’ or the ‘optimal’) regression model from which the impacts of independent variables are inferred. This note is to draw the attention of ecologists to a relatively recent method, hierarchical partitioning, that does not aim to identify a best regression model as such but rather uses all models in a regression hierarchy to distinguish those variables that have high independent correlations with the dependent variable. Such variables are likely to be most influential in controlling variation in the dependent variable. Hierarchical partitioning is not to be regarded as a substitute for experimental manipulation when that is appropriate, but it is likely to produce better deductions than common regression approaches in the many ecological situations in which manipulation is impossible or of doubtful value.     

Timing and sequence of brain activity in top-down control of visual-spatial attention

Recent brain imaging studies using functional magnetic resonance imaging (fMRI) have implicated a frontal-parietal network in the top-down control of attention. However, little is known about the timing and sequence of activations within this network. To investigate these timing questions, we used event-related electrical brain potentials (ERPs) and a specially designed visual-spatial attentional-cueing paradigm, which were applied as part of a multi-methodological approach that included a closely corresponding event-related fMRI study using an identical paradigm. In the first 400 ms post cue, attention-directing and control cues elicited similar general cue-processing activity, corresponding to the more lateral subregions of the frontal-parietal network identified with the fMRI. Following this, the attention-directing cues elicited a sustained negative-polarity brain wave that was absent for control cues. This activity could be linked to the more medial frontal–parietal subregions similarly identified in the fMRI as specifically involved in attentional orienting. Critically, both the scalp ERPs and the fMRI-seeded source modeling for this orienting-related activity indicated an earlier onset of frontal versus parietal contribution (∼400 versus ∼700 ms). This was then followed (∼800–900 ms) by pretarget biasing activity in the region-specific visual-sensory occipital cortex. These results indicate an activation sequence of key components of the attentional-control brain network, providing insight into their functional roles. More specifically, these results suggest that voluntary attentional orienting is initiated by medial portions of frontal cortex, which then recruit medial parietal areas. Together, these areas then implement biasing of region-specific visual-sensory cortex to facilitate the processing of upcoming visual stimuli.     

Bayesian nonparametric inference for panel count data with an informative observation process

In this paper, the panel count data analysis for recurrent events is considered. Such analysis is useful for studying tumor or infection recurrences in both clinical trial and observational studies. A bivariate Gaussian Cox process model is proposed to jointly model the observation process and the recurrent event process. Bayesian nonparametric inference is proposed for simultaneously estimating regression parameters, bivariate frailty effects, and baseline intensity functions. Inference is done through Markov chain Monte Carlo, with fully developed computational techniques. Predictive inference is also discussed under the Bayesian setting. The proposed method is shown to be efficient via simulation studies. A clinical trial dataset on skin cancer patients is analyzed to illustrate the proposed approach.     

Aortic valve stenosis: an active atheroinflammatory process

PURPOSE OF REVIEW: To summarize the current understanding of the pathobiology of aortic valve stenosis and portray the major advances in this field. RECENT FINDINGS: Stenotic aortic valves are characterized by atherosclerosis-like lesions, consisting of activated inflammatory cells, including T lymphocytes, macrophages, and mast cells, and of lipid deposits, calcific nodules, and bone tissue. Active mediators of calcification and cells with osteoblast-like activity are present in diseased valves. Extracellular matrix remodeling, including collagen synthesis and elastin degradation by matrix metalloproteinases and cathepsins, contributes to leaflet stiffening. In experimental animals, hypercholesterolemia induces calcification and bone formation in aortic valves, which can be inhibited by statin treatment. The potential of statins to retard progression of aortic valve stenosis has also been recognized in clinical studies; however, further prospective trials are needed. Angiotensin II-forming enzymes are upregulated in stenotic valves. Angiotensin II may participate in profibrotic progression of aortic valve stenosis and may serve as a possible therapeutic target. SUMMARY: Recent findings regarding the interaction of inflammatory cells, lipids, mediators of calcification, and renin-angiotensin system in stenotic valves support the current opinion of aortic valve stenosis being an actively regulated disease, potentially amenable to targeted molecular therapy. Evidence from prospective clinical studies is eagerly awaited.     

Biogenesis as an evolutionary process

Summary The earliest fossil stromatolites present evidence of a complex ecosystem of photosynthetic organisms. Because the origin of present life can be dated within a few hundred million years prior to these fossils, their complexity poses a problem. A heuristic model outlines the first radiation leading to the universal ancestor.     

Convergence as an evolutionary process

Convergent evolution may be elucidated as a series of stages which show a decreasing number of common characters and become more and more generalized, the characters defining the stages finally being difficult to interpret as adaptive. The series starts with ecological races (genoecodemes), proceeds with parallel ecological races (ecotypes) to sub-life-forms, life-forms (growth forms), ecophysiological 'types' (e.g. xerophytes, halbphytes), and finishes with morpho-physiological levels. The ecotype concept is discussed, and it is proposed that it be restricted to cover parallel ecological races. Other discussions concern the possible occurrence of neutral or non-adaptive characters, the bunching of character gradients, and the two main classification systems of living entities.     

Implementing Gaussian process inference with neural networks

Gaussian processes compare favourably with backpropagation neural networks as a tool for regression, and Bayesian neural networks have Gaussian process behaviour when the number of hidden neurons tends to infinity. We describe a simple recurrent neural network with connection weights trained by one-shot Hebbian learning. This network amounts to a dynamical system which relaxes to a stable state in which it generates predictions identical to those of Gaussian process regression. In effect an infinite number of hidden units in a feed-forward architecture can be replaced by a merely finite number, together with recurrent connections.