Cosyne 2009 Workshops
March 3, 2009
Snow Bird, Utah
Workshop Title
Common computational principles of attention and decision making: can they account for unexpected observations in parietal cortex?
Organizer(s)
Anne K. Churchland, University of Washington, Seattle
Jochen Ditterich, University of California, Davis
Justin L. Gardner, New York University
Abstract
Current knowledge about the physiological bases of attention and decision-making has, in large part, arisen from opposing experimental tactics. Isolating behavioral or neuronal modulations due to attention requires ruling out stimulus-related artifacts. That is, when studying attention, sensory evidence should be held constant while priors (e.g. where the target is located, or what color it is) are modified. Conversely, studying decision-making mechanisms requires understanding the transformation of sensory signals into decisions and is best understood by controlling for potential artifacts due to changing behavioral states. Thus sensory evidence is modified (e.g. stimulus strength) while priors are held constant. While inquiry into attention and decision making start from opposite experimental paradigms, they both seek to find and test computational principles that link behavior to its underlying physiology.
Nowhere in the brain is this dichotomy of approach as evident as in the parietal cortex. Starting from the observation of spatial neglect after parietal lesions, there is a long tradition that views the parietal cortex in terms of its proposed role in directing spatial attention. However, much recent work has implicated parietal cortex in a number of other behaviors in addition to attention. For example, hallmarks of decision–making can be seen in parietal cortex, and recent evidence suggests it may even play a role in determining quantity and discriminating shapes as well. To what extent this diversity is simply due to differences in experimental and theoretical techniques that mirror the dichotomy of approach towards attention and decision-making is unclear.
This workshop will begin with an overview of recent work in parietal cortex to highlight the diversity of neural responses that are observed in this part of the brain. Next, theorists and experimentalists will attempt to find unifying computational principles for the diverse behavioral and neurophysiological observations, with an emphasis on attention and decision-making. A combined theoretical framework that can account for all observed phenomena simultaneously should form the foundation of our understanding of the physiological basis for attention and decision-making. With such a framework, perhaps many of the diverse experimental findings in parietal cortex can also be unified into evidence for a common set of computations performed by cortical neurons.
Speakers
Morning
08:00-08:10 AM Introduction - Jochen Ditterich (University of California, Davis): The functional and computational role of parietal cortex
In this introduction I will give a historical overview over some experimental and clinical findings that provide insight into the functional role of parietal cortex. The goal is to set the stage for a discussion of how some recent, unexpected experimental findings in parietal cortex might fit into a unifying concept of parietal function and computation.
08:10-08:40 AM Peter Janssen (Katholieke Universiteit Leuven, Belgium): Shape processing in posterior parietal cortex
The analysis of shape is critical for object recognition, saccade planning and grasping. We investigated the neural representation of 2D- and 3D shape in areas LIP and AIP, respectively. Most LIP neurons were significantly tuned to 2D shape but - unlike inferotemporal neurons – showed weak size and position invariance. Many LIP neurons displayed spurious shape selectivity arising from accidental interactions between stimulus and receptive field. Although many LIP neurons showed spatially-selective saccadic activity, the receptive field for saccades did not always correspond to the receptive field for shapes. In AIP, robust selectivity for curved surfaces was observed at relatively short latencies. Unlike inferotemporal neurons, AIP neurons tolerated the presence of disparity discontinuities in the stimulus but provided reliable information on the degree of curvedness of the stimulus. Overall, the shape representation in the dorsal stream appears to differ radically from the known representation of shape in the ventral stream.
08:40-08:50 AM Questions and Discussion
08:50-09:20 AM David Freedman (University of Chicago): Neuronal encoding of the behavioral significance of visual stimuli in parietal cortex
We have an impressive capacity to recognize the behavioral significance, or category membership, of a wide range of sensory stimuli. This ability is critical because it allows us to respond appropriately to stimuli that we encounter in our interactions with the environment. Recently, we recorded from lateral intraparietal (LIP) neurons during a categorization task in which 360º of motion directions were grouped into two arbitrary categories that were divided by a learned category boundary. These recordings revealed that LIP neurons robustly encoded stimuli according to their learned category membership, suggesting that parietal visual representations can reflect abstract information about the learned significance of visual stimuli. More recent work has also revealed strong neuronal encoding of the learned associations between pairs of static visual-shapes in LIP, suggesting that parietal cortex likely plays a role in encoding the significance of both spatial (e.g. motion or space) and non-spatial (e.g. shape) stimuli.
09:20-09:30 AM Questions and Discussion
09:30-09:40 AM Coffee break
09:40-10:10 AM Anne K. Churchland (University of Washington, Seattle): Exploiting the LIP salience map to study neural mechanisms of decision-making.
Recording studies in the lateral intraparietal area (LIP) have interpreted neural activity as relating to (among other things) motor planning, visual attention, expected value of a reward, elapsed time, stimulus shape and perceptual decision-making. Rather than fuel the debate about which internal state or stimulus variable is encoded in LIP, I will interpret LIP responses in terms of what aims to be a bipartisan expression: their reflection of the salience of different parts of the visual field. I will first argue that certain tasks exploit the salience map to uncover mechanisms of decision-making. For example, the value of the threshold or bound that terminates a decision or the relative changes in firing rates that accompany weak versus strong motion. Next, I will ask under what conditions it is possible to make use of the salience map to understand decision-making mechanisms. The key feature, perhaps, is that salience is conferred gradually to a distinct location in space. Lastly, I will present evidence that the spatial configuration of the salience map depends on the context of the task at hand. In the context of a motion discrimination task, this argues that the spatial configuration of the salience map is not dictated solely by the direct anatomical connections between MT and LIP.
10:10-10:20 AM Questions and Discussion
10:20-10:50 AM Jacqueline Gottlieb (Columbia University): Distinct but interacting attentional and decisional signals in area LIP
The lateral intraparietal area (LIP) has been implicated in attention and decision-making. These functions have been studied by independent laboratories with different experimental paradigms, and so far there has been scant discussion about how they relate to each other: are attention and decision identical, distinct but interacting, or mutually exclusive neural and psychological processes? Here I argue that attention and decision are distinct but overlapping forms of selection, and that they jointly influence LIP activity. Attending and deciding both involve the selection of a task-relevant object or motor act. In the case of spatial attention and oculomotor decisions, selection is spatial; thus, associated neural activity must be spatially tuned and higher for targets relative to non-target objects. These are indeed consistent hallmarks of LIP activity across different laboratories. However, attention and decisions differ in the underlying purpose of the selection. A “decision” is the selection of an action that is closely related to an outcome. In a typical experiment, subjects may choose to make an eye movement to one of several targets, or to press one of several buttons, and are given immediate feedback (reward or penalty) regarding their choice. “Attention” on the other hand, is the selection of a source of information that is relevant for the action. Subjects must attend to the relevant information in order to make the correct decision; however, they are not rewarded for a specific deployment of attention, but are only rewarded if their attentional choice led to the appropriate decision. Thus attention may be viewed as a second-order, indirect decision: the decision of which source of information is most likely to lead to an action that leads to reward. I will discuss evidence from our laboratory showing that 1) LIP neurons primarily reflect attentional selection, 2) neurons show additional decisional signals that modulate their attentional response, 3) attentional signals are modulated by reward expectation in ways that are not always consistent with optimal decision-making, and 4) neurons undergo reward-dependent plasticity. Taken together these findings suggest that LIP reflects a process by which the brain learns to identify and select valuable sources of information and link them with the appropriate action.
10:50-11:00 AM Questions and Discussion
Afternoon
04:30-04:40 PM Introduction - Justin L. Gardner (New York University)
In the afternoon session, we will move from the parietal cortex into a more general discussion of empirical and theoretical aspects of attention and decision making with an eye towards common computational principles that underlie these two important fields. Perhaps a unified framework for understanding attention and decision making may help to understand the diversity of recent physiological findings in the parietal cortex, which will have been showcased in the morning session.
04:40-05:10 PM David Heeger (New York University): Normalization Model of Attention
Attention has been found to have a wide variety of effects on the responses of neurons in visual cortex. John Reynolds (Salk) and I have developed a model of attention that exhibits each of these different forms of attentional modulation, depending on the stimulus conditions and the spread (or selectivity) of the attention field in the model. The model helps reconcile proposals that have been taken to represent alternative theories of attention. We argue that the variety and complexity of the results reported in the literature emerge from the variety of empirical protocols that were used, such that the results observed in any one experiment depended on the stimulus conditions and the subject’s attentional strategy, a notion that we define precisely in terms of the attention field in the model, but that has not typically been completely under experimental control.
05:10-05:20 PM Questions and Discussion
05:20-05:50 PM James Mazer (Yale University): Top-down spatial attention: retinotopic or spatiotopic?
What happens to a sustained locus of spatial attention when we move our eyes? Does the region of behavioral enhancement shift with the eyes (retinotopic) or does it remain fixed in the external visual environment (spatiotopic)? The answer to this question depends in part on the exact nature of the behavioral task. However, even when the task requires subjects to encode and maintain the spatiotopic location of a cue, we observe behavioral benefits (both in accuracy and reaction time measures) at unattended retinotopic locations. These psychophysical results are consistent with a retinotopically organized salience map playing a key role in the maintenance of spatial attention. Results from an fMRI experiment using a novel gaze-contingent display paradigm show evidence of retinotopic BOLD activity in the ventral visual processing stream of human observers consistent with this hypothesis.
05:50-06:00 PM Questions and Discussion
06:00-06:10 PM Coffee break
06:10-06:40 PM Paul Cisek (University of Montreal):Attention and action selection through a distributed consensus
Evolution, or descent-with-modification, does not produce entities without a shared history. This applies to functional entities as it does to species, and concepts we may label as attention or decision might both have descended from the same selection functions that were central to the behavior of our ancestors. As many have suggested, attention and decision-making continue to play related and overlapping roles in behavior, and appear to involve similar neural mechanisms of biased competition. In my talk, I will describe a model in which attentional and decisional variables are commingled in cortical activity, and the selection of action emergences through a distributed consensus that involves a large portion of the fronto-parietal cortex and the cortico-striatal system. Such a model predicts that neural correlates of almost any factor which influences action selection will appear in parietal and premotor cortex, appearing as attention from one experimental perspective and a decision from another.
06:40-06:50 PM Questions and Discussion
06:50-07:20 PM Xiao-Jing Wang (Yale University):Recurrent circuit dynamics underlying decision making and selective attention
07:20-07:30 PM Questions and Discussion
