Memory, Attention, Decision (MAD) Lunch


Memory, Attention & Decision Lunch was founded as an effort to promote interdisciplinary discussions and collaborations across leading labs investigating how the brain learns and remembers, pays attention, and makes decisions. The motivation is to bring into dialog diverse labs who study similar brain regions (specifically fronto-parietal cortex) from different perspectives. We discuss neural mechanisms of memory, attention, & decision-making from monkeys to humans.

We hold monthly meetings that bring together 10 Stanford labs (and counting). Please join us!

If you’d like to be included in the email list for MAD lunch, please email me.

Our current meeting schedule is the 3rd Tuesday of each month, noon - 1:30p (unless otherwise specified)



3/24/15: Prof Brian Knutson and Josiah Leong

Coherence of a tract connecting anterior insula to nucleus accumbens mitigates positive skew preference

Functional neuroimaging research has implicated the nucleus accumbens and the anterior insula in financial risk taking. Here, we document a novel tract connecting the anterior insula to nucleus accumbens, demonstrate that its coherence is associated with reduced susceptibility to positive skewed gambles, and determine that this association is mediated by reduced functional activity in the nucleus accumbens. 

3/24/15: Alex Huth of Jack Gallant’s lab (UC Berkeley)

Mapping semantic representation in the brain using natural language

Human beings have the unique ability to extract the meaning, or semantic content, from spoken language. Yet little is known about how the semantic content of everyday narrative speech is represented in brain. We used a new fMRI-based approach to show that semantic information is represented in complex cortical maps that are highly consistent across subjects. Using BOLD data collected while subjects listened to several hours of natural narrative stories, we constructed voxel-wise semantic regression models that accurately predict BOLD responses based on semantic features extracted from the stories. These semantic features were defined using a statistical word co-occurrence model. We then used a novel Bayesian generative model of cortical maps to discover how the representations revealed by voxel-wise modeling are organized across the cortical sheet. The results of these analyses show that the semantic content of narrative speech is represented across parietal cortex, prefrontal cortex, and temporal cortex in complex maps comprising dozens of semantically selective brain areas.

2/17/15: Desmond Oathes of Amit Etkin’s lab

Probing fear memory circuitry with concurrent TMS-fMRI

Non-human anatomical tracing has revealed largely uni-directional inputs from ventrolateral PFC (vlPFC) to amygdala. These areas often co-activate in human neuroimaging studies of emotion but, to date, no causal evidence of communication between these brain areas has been established in humans. To determine the directional causal influence of vlPFC activation on the amygdala and to investigate possible pathway abnormalities in posttraumatic stress disorder (PTSD), we stimulated both right and left vlPFC in patients with PTSD as well as trauma exposed healthy participants interleaved with fMRI acquisitions.  Our findings support a functional pathway in humans similar to that found in animal work and suggest a specific pathway abnormality in PTSD patients.

2/3/15: Chris Gorgolewski of Russ Poldrack’s lab - a new online tool for cognitive neuroscientists is a new web repository for sharing statistical maps of the human brain. In this brief presentation I will explain how it can be used to improve data exchange between collaborators as well as perform formal reverse inference giving you more insight into your results. I will dive deeper into how the database as a whole can be used for performing powerful meta-analyses thus potentially increasing the impact of your research.

12/15/14: Becket Ebitz of Tirin Moore’s lab

Target selectivity in the frontal eye field (FEF) is blunted during exploratory choice

In order to maximize reward intake within the uncertain and variable environments typical of nature, decision-makers must balance two goals. They must exploit known-rewarding targets, but also explore alternative targets in order to reduce uncertainty about the environment and discover targets which may be more rewarding than the currently favored options. It is unclear how these distinct goal states are expressed by the brain, though two distinct hypotheses have been proposed. First - exploration may enhance attention, relative to exploitation, in order to facilitate learning. Alternatively, exploration may enhance decision noise, relative to exploitation, in order to make non-preferred targets more likely to be chosen. These two alternative hypothesis make competing predictions about how these goal states should shape target-selective activity in the frontal eye field (FEF), a prefrontal oculomotor-control region. Target selectivity is the enhanced neuronal response to one stimulus at the expense of other, simultaneously presented stimuli. Target selectivity is a hallmark of activity in FEF and is associated with both saccadic and attentional selection. However, here we find that target selectivity is largely absent in FEF during exploratory decision-making states, consistent with the hypothesis that exploration results in increased decision noise but not consistent with an attentional selection account. Noisy decision rules are a computationally efficient way to induce exploration in artificial decision-makers and these observations suggest that the brain may have arrived at a similar solution. Moreover, our results suggest that target selectivity in FEF is not a deterministic consequence of choice. Instead, it may depend on the exploitative states so essential for the performance of most laboratory tasks.

11/24/14: Prof Brian Knutson

Microlending mechanisms: From neural activity to aggregate behavior

10/20/14: Prof Mark Seidenberg from University of Wisconsin-Madison

How does reading work and how can we tell? Or, Who is Max Coltheart and why is he saying those things?

6/10/14: Jacek Dmochowski of Tony Norcia’s lab

Reliable components of EEG are the neural signatures of accumulation-to-bound

in a fine perceptual decision-making task

Perceptual decision-making has been extensively studied with an accumulation-to-bound model, in which evidence for two candidate choices is simultaneously integrated until the evidence for one of the alternatives reaches a set decision-bound. The neural correlates of accumulation-to-bound have been previously found in monkey area LIP, where spiking activity follows a stereotyped ramping trajectory whose slope and duration predict the choice and reaction time (RT) of the monkey.   Here I will show that appropriately-extracted human electroencephalographic (EEG) signal components exhibit the characteristic features of accumulation-to-bound in a fine motion-direction discrimination paradigm.  Specifically, we compute projections of the data which exhibit maximal trial-to-trial covariance, thus capturing the neural activity reliably evoked by our paradigm.  The two most reliable components (RCs) are located over medial parietal and premotor cortices, respectively, and exhibit ramping time-courses which are strongly modulated by task difficulty (RC1 only) and reaction time (RC1 and RC2).  These results provide a framework for studying perceptual decisions in human, where we can observe multiple accumulation processes unfolding in parallel on a single-trial basis.  To conclude, I will summarize our preliminary attempts at modulating task RT with transcranial direct current stimulation (tDCS).

5/14: Break

3/13/14: Diogo Peixoto of Bill Newsome’s lab

Neural correlates of Decision Formation in PMd in a Perceptual Discrimination Task

Many prior studies of visually based decision-making have employed stimuli in which subjects decide the average direction of motion in stochastic random dot stimuli.  Because subjects accumulate sensory evidence over time before committing to a decision, this framework provides an opportunity for electrophysiological analysis of evolving neural decision variables that reflect integrated evidence.  Most prior studies, however, have employed traditional single unit recording techniques, which limits their power to assess the dynamics of evidence accumulation and decision formation on single trials.

To address these issues, we recorded simultaneously from multiple units during the random dots motion discrimination task, using arm reach movements as the operant response.  Neural activity was recorded from dorsal premotor (PMd) cortex, an area that is easily accessible to ‘Utah’ multielectrode arrays.  Our goals were twofold: 1) to determine whether PMd activity reflects evidence accumulation toward decision about arm movement target selection, analogous to well-studied processes in pre-oculomotor structures, and 2) to leverage the statistical power conferred by simultaneous recordings to obtain insights into the decision formation process on single trials.

We used logistic regression to predict the monkey’s choices from neural population activity on individual trials using leave-one-out cross-validation.  Logistic predictions were made in a 150 msec sliding window, allowing us to observe the evolution of decision-related activity in PMd.  For strong motion coherences, average predictive activity, as measured by the fraction of correctly predicted trials, exceeded chance levels ~ 200 msec after onset of the visual stimulus, and reached 80% correct by 300 msec. As in LIP and other pre-oculomotor structures, predictive activity varied with stimulus coherence, rising faster and reaching higher levels for stronger coherences.

Building on these results we went one step further and explicitly searched for signals in the population activity that covary with task variables of interest.  In our study these variables are the animal's choice at the end of the trial and the coherence of the stimulus. As expected from the logistic regression results we found a strong choice-related signal that emerges ~200 msec after the onset of the dots and lasts for the duration of the trial. In addition we also found an orthogonal stimulus coherence signal, that reliably separates the different stimulus conditions but only for a short time window during the dots presentation.   This signal can be interpreted as a correlate of momentary motion evidence, since it reflects the stimulus strength but doesn't affect choice.

Together, our data suggest that PMd activity reflects the accumulation of sensory evidence in the reach system, providing an opportunity to examine single trial neural dynamics underlying decision formation.

2/13/14: Chandramouli Chandrasekaran from Krishna Shenoy’s lab

Neural correlates of action choice and RT in dorsal premotor cortex

How does the primate brain make decisions based on sensory input and decide which action to perform when? Lesion and electrophysiological studies suggest a role for dorsal premotor cortex (PMd) in mapping sensory cues to motor actions. We studied the neural dynamics in dorsal premotor cortex during a task involved in mapping an ambiguous sensory cue to movement. In particular, we investigated whether activity in PMd was consistent with a candidate decision variable?

A trained monkey used his right arm to report the dominant color in a central static checkerboard composed of isoluminant red and green squares. The percentage of red and green in the stimulus varied from trial to trial. The monkey’s behavior was typical of visual RT tasks; increases in difficulty led to more discrimination errors and slower RTs. Most RTs varied between ~400 ms for the easiest discrimination to ~550 ms for the hardest discriminations (range: 300 to 1000 ms). While the monkey performed this task, we recorded the activity of single neurons and small populations of neurons (using laminar multi contact probes) from the arm region of left PMd.

Responses in PMd as a function of checkerboard difficulty suggested a candidate decision variable. When aligned to checkerboard onset, activity differences for left vs right choices increased faster for clearer checkerboards and slower for ambiguous checkerboards. In addition, at the time of movement onset, activity differences were largely the same across different checkerboard difficulties.

Examining neural responses on the basis of RT revealed a finer grained categorization of neural responses in PMd and reinforced the presence of a candidate decision variable. As expected, one subset of the population was only movement sensitive, diverging in activity for different choices only on or around (~200 ms) the reach onset. However, another population showed an interesting response profile. For these cells, immediately after stimulus onset, cells selective for eventual right and left reaches exhibited comparable firing rates. Over time, however, the absolute difference in activity of these two populations increased; achieving on average a consistent 25 spikes/s difference (~100 ms) prior to movement onset. Importantly, when aligned to movement onset, the evolution of this activity difference was faster for fast RTs and slower and earlier for slower RTs. Furthermore, these effects were present within a checkerboard difficulty and thus not a spurious result of independent correlations between task difficulty and RT and task difficulty and neural responses. These results suggest that at least the activity of a subpopulation of PMd cells is consistent with a candidate decision variable.

Finally, on a trial-by-trial basis, firing rates of a small population of simultaneously recorded neurons could be used to decode the choice of the animal well before movement onset. Again, a variation with RT was observed --- eventual choice could be decoded earlier for slower RTs. Taken together, these results suggest a candidate decision variable may be reflected in PMd activity and may mediate the selection of the action to perform and when to perform it.

12/9/13: Bernhard Staresina from Anthony Wagner’s lab

Episodic memory in the medial temporal lobe

In this talk I will discuss the contributions of different medial temporal lobe (MTL) subregions to episodic memory. First, I will show imaging data that reveal the division of labor in the MTL for associative memory formation. In the second part, I will use intracranial EEG data to discuss the temporal dynamics among these MTL regions. Finally, I will show data that capture the reinstatement of past representations during both retrieval and during post-learning offline periods.

11/18/13: Franco Pestilli from Brian Wandell’s lab

Model-based neuroanatomy: Validation and statistical inference in living connectomes

Magnetic resonance diffusion imaging and computational tractography are the only technologies that enable neuroscientists to measure white matter in the living human brain. In the decade since their development, these technologies revolutionized our understanding of the importance of the human white-matter for health and disease. There are good reasons to make these measurements in human. The human brain (1400 g) is 15 times the volume of the rhesus monkey (90 g), 700 times the volume of the rat (2 g) and 2,300 times the volume of the mouse brain (0.6 g). The human brain comprises of functionally specific clusters of maps communicating via an extensive network of long-range, myelinated, axonal projects. The size of the human brain imposes significant challenges for communicating across different regions

Prior to these technologies, the white matter was thought of as a passive cabling system. But modern measurements show that white matter axons and glia respond to experience and that the tissue properties of the white matter are transformed during development and following training. The white matter pathways comprise a set of active wires and the responses and properties of these wires predict human cognitive and emotional abilities in health and disease. We can now predict confidently that to crack the neural code in mapping the human brain, neuroscientists will have to develop an account of the connections and tissue properties of these active wires.  Whereas there are many impressive findings, it is widely agreed that there is an urgent need to keep developing and improving tractography methods.  The need for a systematic approach to tractography validation and for a framework to perform statistical model testing can be seen in recent reports in Science that set out to characterize human white matter structure.

I will present new methods to perform both tractography validation and statistical hypotheses testing on the network of brain connections.  These new methods improve current techniques in fundamental ways and can be applied to any type of diffusion data.  I will show that by using the methods we were able to identify a major white-matter pathway communicating information between the dorsal and ventral visual streams, the Vertical Occipital Fasciculus (VOF). This pathway is large and its organization suggests that the human ventral and dorsal visual streams communicate substantial information through areas V3A/B and hV4/VO-1. We suggest that the VOF is crucial for transmitting signals between regions that encode object properties including form, identity and color information and regions that map spatial location to action plans.

9/17/13: Brandon Turner from Jay McClelland’s lab

Extensions of a Dynamic, Stimulus-Driven Model of Signal Detection

Signal detection theory forms the core of many current models of cognition, including memory, choice,and categorization. However, the classic signal detection model presumes the a priori existence of fixed stimulus representations -- usually Gaussian distributions -- even when the observer has no experience with the task. Furthermore, the classic signal detection model requires the observer to place a response criterion along the axis of stimulus strength, and without theoretical elaboration, this criterion is fixed and independent of the observer’s experience. We present a dynamic, adaptive model that addresses these two long-standing issues. Our model describes how the stimulus representation can develop from a rough subjective prior and thereby explains changes in signal detection performance over time. In the talk, we will discuss new efforts to extend the model to include response time and confidence in a two-choice detection task as well as a four-choice classification task. We present simulations of the model to examine its behavior and several experiments that provide data to test the model.

8/13/13: Summer break

7/8/13: Ian Ballard from Sam McClure’s lab

Dynamic Causal Modeling (DCM) methods

In recent years, there has been a growing interest in computational techniques that measure connectivity across entire brain networks. I will discuss one such technique called Dynamic Causal Modeling (DCM). DCM aims to infer both the (directed) connectivity structure of a network of regions as well as how the strength of these connections varies in response to experimental manipulations. I will give an overview of the theory behind DCM and demonstrate one of the ways in which it can be used by discussing the analysis pipeline and results of:

Dorsolateral Prefrontal Cortex Drives Mesolimbic Dopaminergic Regions to Initiate Motivated Behavior;31/28/10340

5/21/13: Kendrick Kay (Brian Wandell’s lab) and Karen LaRocque (Anthony Wagner’s lab)

Critiques of MVPA methods

Kendrick and Karen will discuss issues related to MVPA methods, recently raised in the following papers: 

Confounds in multivariate pattern analysis: Theory and rule representation case study

Searchlight analysis: Promise, pitfalls, and potential

4/9/13: Sara Szczepanski from Bob Knight’s lab at UC Berkeley

Dynamic frontal-parietal interactions facilitate spatial attention:

Evidence from electrocorticography (ECoG)

Attention is a critical component of visual perception and goal-directed behavior that allows the brain to allocate its limited resources depending on current task demands. A number of frontal and posterior parietal cortical (PPC) areas, often referred to collectively as the fronto-parietal attentional control network, seem to be crucial for controlling the attentional selection process in both humans and non-human primates. This fronto-parietal attentional control network includes the lateral prefrontal cortex (LPFC), the frontal eye fields (FEF), and the supplementary eye field (SEF) in frontal cortex and the superior parietal lobule (SPL) and portions of the intraparietal sulcus (IPS) in PPC. Although a multitude of studies have examined the functions of this network in the human brain using various neuroimaging techniques, considerably less is known about how these frontal and parietal areas interact dynamically to produce behavior on a sub-second temporal scale.

We examined the temporal dynamics and interactions within and between regions of the human fronto-parietal network and visual cortex using electrocorticography (ECoG). ECoG signals were measured directly from subdural electrode arrays that were implanted in patients undergoing intracranial monitoring for localization of epileptic foci, allowing for exceptional spatial and temporal resolution. Subjects (n=7) performed a dynamic reaction time task, in which they allocated visuospatial attention to either the right or left visual field and responded when a target eventually appeared somewhere in the visual field.

In each individual subject, we found increases in cross-frequency coupling (CFC) between high gamma power (70-180 Hz) and delta/theta phase (2-5 Hz) within electrodes over frontal, parietal, and occipital cortex during allocation of spatial attention, which was attenuated in the unattended condition. In addition, we found significant increases in phase coherence in the delta (2-4 Hz) and theta (5-8 Hz) frequency bands between intrahemispheric frontal, parietal, and visual electrodes that was stronger when subjects allocated attention towards the contralateral (vs. ipsilateral) visual field. The increases in coupling across frontal-parietal and visual areas tracked attentional behavioral performance (reaction time) on a trial-by-trial basis.  These results highlight the roles of phase-amplitude CFC and phase coherence as selective mechanisms for communication within and between human fronto-parietal and visual areas, which adjust parameters on a sub-second basis depending on momentary attentional demands.

2/11/13: Prof Anne Sereno from University of Texas, Houston

Executive function, tablets, and population coding

Prefrontal cortex is thought to be critical for executive functions such as attention and working memory. Executive dysfunction is characteristic of many human disorders and can be important in distinguishing subtypes, tracking disease status or progression, and evaluating treatment effects.  We briefly discuss the advantages and disadvantages of eye tracking and present a novel iPad-based approach to measure executive function. Prefrontal cortex receives projections from ventral and dorsal visual streams thought to be important for object and spatial processing, respectively.  Recent work suggests a blurring of properties in these two cortical streams.  We discuss how a novel, "intrinsic" approach to population decoding may better characterize functional differences in ventral and dorsal streams, and eventually lead to better characterization of the prefrontal areas to which they project.


1/15/13: Michael Waskom from Anthony Wagner’s lab

Contextual representations in prefrontal and parietal cortex make

distinct contributions to cognitive control

The ability to flexibly allocate attention to different aspects of the visual environment and respond to stimuli in a contextually appropriate manner is thought to involve a network of regions in frontal and parietal cortex. Specifically, it has been proposed that distributed contextual representations in frontoparietal areas serve to enhance goal-congruent processing in sensorimotor regions and thus enable adaptive behavior. We used linear classifier decoding analyses of fMRI data in regions derived from intrinsic functional networks to understand how contextual information is represented across cortex and what role those representations play in the top-down control of flexible behavior. Although we found that both lateral frontal and parietal areas transiently encode task context during behavior, their functional roles could be dissociated such that parietal cortex was more directly involved in enhancing relevant sensory information while processing in the prefrontal cortex was more closely related to making an effective response.

11/27/12: Hiroyuki Nakahara from RIKEN Brain Science Institute

Learning to simulate and predict the other

Learning and predicting others’ minds is critical for social cognition, but how it is done remains largely unknown. According to theories in social cognition, a simple conception is that humans simulate others’ mental process by directly recruiting one’s own process to model others’ minds. Using human fMRI with model-based analyses based on frameworks of reinforcement learning and value-based decisions making, we found that simulation involves two hierarchical learning signals: a reward prediction error, generated by simulation of direct recruitment to model others’ valuation and encoded in ventromedial prefrontal cortex, and an action prediction error, based on simulation and observation of the other’s choices to track others’ variability and encoded in dorsolateral/dorsomedial prefrontal cortex. These findings show that humans can learn to predict others’ minds from simulation, using a scaffold of mentalizing signals.

Cf. Suzuki et al Neuron (2012). Nakahara & Hikosaka Neuroscience Research (in press).

11/6/12: Christian Rodriguez of Sam McClure’s lab

An integrated frontoparietal network underlying value accumulation in intertemporal choice

Drift diffusion models (DDMs) account for behavior and neural activity in numerous judgment and decision scenarios. I will discuss the results of a recent study that extends the applicability of these models by showing that a DDM provides accurate fits in an intertemporal choice task in which participants are asked to choose between immediate and delayed monetary rewards. Using high-density electroencephalography (EEG), we found that accumulated value signals obtained from best fitting DDM parameters correlated with event related potential (ERP) amplitude over anterior cingulate cortex (ACC). This ACC site was a hub for a fronto-parietal network that also included dorsolateral prefrontal (DLPFC) and posterior parietal cortex (PPC). Synchronous activity within this network, quantified by a phase-locking statistic, predicted faster and more consistent decisions. Moreover, ERPs over PPC were sensitive to value information and preceded ACC activity. On the basis of these data, we propose a model of intertemporal choice in which value-maximizing choices depend on distributed synchronous activity between frontoparietal regions.

10/9/12: Juan Gao of Jay McClelland’s lab

Decision Making +

Perceptual decision making can be thought of as the process through which we use sensory information to guide behavior, and it is a ubiquitous part of life.  However, real life is far beyond just an arrow from stimulus to action.  In real life situations, decision makers constantly need to consider multiple factors, and decision making process constantly integrates, coordinates and even interacts with other processes such as sensing and acting. In this presentation, I will address some of the issues that we need to consider in order to approach real-life problems.  I will focus on the integration of decision making and reward, decision making and action. Given time, I will also talk about decision making with multimodal sensory information.

9/18/12: Prof Stefano Baldassi

Overt and covert remapping of visual information at the time of saccades:

Hints from psychophysical classification images

We actively scan our environment with fast ballistic movements called saccades, which create large and rapid displacements of the image on the retina. At the time of saccades, vision becomes transiently distorted in many ways: briefly flashed stimuli are displaced in space and in time, and spatial and temporal intervals appear compressed. In my talk I will report a series of studies exploiting the psychophysical technique of classification images to study the dynamics of visual mechanisms during saccades. In different experiments, we have investigated the spatiotemporal dynamics of overtly attended targets and whether similar dynamics involve also covertly attended visual targets, away from the saccadic pathway. Using the assumption-free classification images technique, we think we achieved original insights mechanisms of stability of the visual world across consecutive eye movements.

8/14/12: Prof Jay McClelland

Rapid Consolidation of Schema Consistent Knowledge: What makes something consistent with a Schema?

Recent work from the lab of Richard Morris (Tse et al, 2007, Science 316: 76; 2011, Science 333:895) has suggested that new information consistent with a well-established 'schema' can be consolidated rapidly into the neocortex.  In the papers describing this work, the authors suggest that their findings pose a challenge to complementary learning systems theory (McClelland, McNaughton & O'Reilly, 1995, Psychol. Rev. 102: 419), which holds that rapid consolidation of arbitrary new information leads to catastrophic interference with existing neocortical memories, thereby motivating the idea that rapid learning of new arbitrary information depends on the hippocampus and other medial temporal lobe structures; these structures then enable a gradual consolidation process, in which the new information is gradually integrated into neocortical networks through an interleaved learning process.

The tension between the new findings and the claims of CLS revolves around the word 'arbitrary' in the passage above.  New simulations using the same examples explored in McClelland, McNaughton, & O'Reilly show that integration of information that is highly consistent with what is already stored in neocortex can occur rapidly and with little interleaving; interleaving is only necessary for inconsistent information.  But there remain interesting questions both at a theoretical and an empirical level.  Just what counts as arbitrary? 

In this talk I will present the details of the relevant experiments from the Morris lab, as well as details of simulations of mine and of a Morris lab researcher, with the goal of encouraging discussion of future theoretical and experimental efforts that might be undertaken to further address the issues.

5/8/12: Prof Mili Milosavljevic

The effects of visual salience vs. preference in consumer attention and choice: Eye-tracking experiments

What drives consumers’ eyes when they are facing a store shelf, and how does this allocation of attention influence their choices?  Do people look at the most colorful, bright, and aesthetically pleasing items, or does their attention go to the most liked and goal-relevant alternatives?  How does attention allocation differ when making fast vs. slower choices? We here examine how bottom-up factors, i.e., visual salience, and top-down factors, i.e., preferences, affect attention allocation during decision-making in a real world setting.  We use (1) the novel computational model of bottom-up dependent attention to determine the effects of bottom-up features, (2) consumer stated preferences to determine top-down effects, (3) eye-tracking to determine where to and for how long people are looking, and (4) both computer generated and natural images of store shelves to study the point-of-purchase context.

4/10/12: Alireza Soltani of Tirin Moore’s lab

Separable influences of reward on prefrontal control of attention and choice behavior

Understanding the neural mechanisms of value-dependent choice and attention have been two of the most important research objectives in systems neuroscience. However, difficulties in dissociating attentional and reward-dependent mechanisms have greatly limited progress in understanding these two processes and how they relate to one another. By combining modeling and experimental approaches, we examined the influence of value on target selection and attention using saccade metrics in a free-choice task. Specifically, monkeys selected between two visual targets, each of which was a drifting grating. It has been shown that saccades directed to such targets are displaced in the direction of visual motion. This “motion induced bias” (MIB) provides an untrained, implicit measure of attentional deployment as it depends on the features of the target and reveals the extent of processing of those features.  We first show that attention is influenced by reward history even when it is not required to obtain reward, and that reward integration for attention differs from that of target selection. Second, we show that FEF microstimulation and reward history exert additive effects on attention, while they interact to guide selection. Third, we show that our experimental observations can be explained by a model in which the efficacy of microstimulation-driven signals and endogenous reward signals interacts competitively.

3/13/12: Vince McGinty of Bill Newsome’s lab

Do you like what you see? The influence of gaze direction on value encoding in orbitofrontal cortex

2/14/12: Prof Brian Knutson

Frontostriatal white matter integrity mediates age-related decline in probabilistic reward learning

Frontostriatal circuits have been previously implicated in reward learning, and emerging findings suggest that both frontal white matter structural integrity and probabilistic reward learning decline with age. This study examined whether declines in human frontostriatal white matter integrity could account for age-related decreases in reward learning in a community adult life-span sample. By combining diffusion tensor imaging (DTI) with a probabilistic reward learning task, we found that (1) learning decreased as a function of age; (2) white matter integrity in pathways running from the thalamus to the medial prefrontal cortex and from the medial prefrontal cortex to the ventral striatum decreased as a function of age; and (3) declines in white matter integrity in these specific paths could statistically account for age differences in learning. The findings suggest that the integrity of ventromedial thalamocorticostriatal white matter pathways critically supports reward learning. These findings also raise the possibility that interventions that can bolster the integrity of frontostriatal pathways might improve learning and decision making.

12/12/11: Miriam Rosenberg-Lee of Vinod Menon’s lab

The alphabet soup of lateral (inferior) parietal cortex: making sense of the structure and function of the AG, SMG, IPL, TPJ, VPC, vPPC, PGa, & PGp

The angular gyrus (AG) is involved in numerous cognitive domains, including language, memory, and mathematics, and is thought to be the lateral nodes of the default mode network. Yet the specific function of this region remains unclear, as does the region’s exact boundaries and even its name. Recently, observer independent cyto-architectonic mapping has revealed two subdivisions within the angular gyrus, the anterior PGa and the posterior PGp regions. Using this more precise anatomic parcellation, we discuss the functional and structural connectivity, task specific activation and the findings of our meta-analysis of co-activation of these two regions. Building on the region’s connectivity to memory and semantic systems, we propose a novel theory of AG function. Specifically, that the AG is crucially involved in linking episodic experiences to semantic knowledge. We discuss this proposal in the context of other theories on the function of this region and potential experiments to test the proposal.

11/21/11: Franco Pestilli of Brian Wandell’s lab

Distinct Effects of Valence and Value on Visual Sensitivity and Cortical Activity

Survival depends on efficient selection and discrimination of stimuli associated with punishment or reward. Efficient value-based selection entails maximizing gain and minimizing loss. But gain-maximization and loss-minimization have opposite emotional valence. Whereas performing to maximize gains has a positive valence, minimizing losses has negative valence. We combined psychophysics and fMRI to separate the effects of valence and value on visual sensitivity and cortical activity in early visual cortex. Behavioral sensitivity improved and cortical activity increased in early visual cortex at the location of a high-value stimulus. Independently of this stimulus valence changed cortical activity in a global, spatially unspecific manner. Cortical activity was higher on negative-valence trials then for positive-valence ones. In contrast to the value effect, the increased activity on negative valence trials did not correlate with a change in visual sensitivity nor was specific for the location of the behaviorally relevant stimuli. We conclude that humans: (1) maximize the chance of obtaining gains and minimize loss by focusing cortical resources to behaviorally relevant high-value stimuli, in a way that is reminiscent of the allocation of spatial attention, (2) brain responses are affected by the positive or negative valence of the stimuli independently of the task at hand. The valence effect is likely of neuromodulatory nature.

10/10/11: Prof Tony Norcia

Sensory Memory and Expectations

9/12/11: Uri Maoz of Cristof Koch’s lab (Caltech)

Neural Prejudice:  

Prestimulus Activity in the Dorsolateral Prefrontal Cortex Predicts Subsequent Value Judgment

Rational value-based decision-making mandates selecting the option with highest subjective expected value, possibly after extensive deliberation. We examined activity in the dorsolateral prefrontal cortex (DLPFC) of monkeys deciding between smaller, immediate rewards and larger, delayed ones. We found that the activity of about a third of neurons predicted the animal’s decision – with respect to the spatial location of the selected target or the more-abstract preferred reward size – even before the decision alternatives were presented. Their predictive power increased as the values of the two options became more similar and the reaction time lengthened. A simple winner-take-all circuit model captured this behavior. It also made several empirically testable predictions about the time-course of the bias-signal and the reaction times, which were borne out by the data. Our data are compatible with the hypothesis that the DLPFC harbors pre-deliberative biases that are later incorporated into the deliberation process during value-based decision-making.

8/22/2011: Matt Kaufman of Krishna Shenoy’s lab

A single-trial view of changes of mind in monkey motor cortex

In order to act, we must make decisions about what action to take. Our choices may be ‘forced’ when only one action is available, or we may have ‘free choices’ when multiple good options present themselves. Moreover, we may acquire additional information mid-decision, prompting a change of mind. Behavioral observations of changes of mind have previously been made by Resulaj et al. (Nature 2009); we examined the covert neural processes that occur during decision-making by combining a novel ‘decision-maze’ paradigm in monkey with large-scale simultaneous neural recordings and recently developed single-trial analytical tools. This combination has allowed us to observe the decision-making process unfold over time on individual trials, providing a window into processes such as vacillation.

7/11/2011: Summer break

6/13/2011: Wouter van den Bos of Sam McClure’s lab

The Value of Victory - Neural Mechanisms of the Winner’s Curse

One of the most interesting but unresolved phenomena in auction behavior is the winner’s curse — the strong tendency for participants to bid more than rational agent theory prescribes, often at a significant loss (Kagel & Levin, 2002). The dominant explanation for the Winner's Curse supposes that people have insufficient cognitive abilities and hence are unable to overcome the curse (e.g. Eyster and Rabin, 2005)

In contrast, we propose that behavior is driven by other factors, such as social values potentially including status (Frank, 1985; Smith, 1990). In a previous behavioral study we showed that the level of social competition in auctions predicted the magnitude of the winner’s curse (van den Bos et al., 2008). Additionally, we observed that the magnitude of the curse typically declined over time, suggesting that bidding decisions are made on the basis of feedback acquired during the auction.

Based on the hypothesis that bidding strategies are the result of (1) feedback-based learning and (2) biases dependent on social context, we developed and employed an extended reinforcement learning model to fit behavioral and neuroimaging data.

In order to determine the nature of the mechanisms that underlie this bidding behavior we simultaneously scanned groups of five participants as they participated in a competitive common value auction. In total, 25 healthy adults participated in this study from which we acquired 22 usable fMRI data sets.

The findings indicate the involvement of neural systems involved in reinforcement learning (ventral striatum) and social cognition (TPJ), supporting the hypothesis that reinforcement learning and social motives drive bidding strategies.

5/9/2011: Sharareh Noorbaloochi of Jay McClelland’s lab

Dynamic representation of reward and stimulus information in decision making process:

An EEG study

In perceptual decision making tasks, prior knowledge about choice payoffs biases decisions toward the higher paying alternative.  However, little is known about how payoff information affects the neural circuits underlying these biased choices in human subjects. In this study, we recorded electroencephalography (EEG) signals from 13 participants while they performed a deadline two-alternative forced choice task. The participants had to detect a horizontal shift in a rectangular stimulus and indicate their choice (left/right) by squeezing one of two dynamometers. On each trial, payoff information was disclosed 1.5 sec prior to stimulus onset. Two randomly-assigned stimulus difficulty levels were employed, and the payoff scheme was randomly assigned to be balanced (equal reward for both alternatives) or unbalanced (one choice worth twice as much as the other). Behaviorally, with unbalanced rewards, fast responses were associated with high reward bias; bias was reduced for slower responses. Neurally, reward bias was observed in the lateralized readiness potential (LRP), a response-related negative-going potential contralateral to the responding hand, in at least two ways: (1) a shift in the baseline activity toward the higher reward alternative prior to stimulus onset, (2) an abrupt rise in activity toward the higher reward alternative 150-200 ms after stimulus presentation, before stimulus information started to affect the LRP. These results suggest that payoff information is reflected in preparatory activity in movement planning structures before sensory information is available. Furthermore, this biased preparatory activity increases quickly once movement is allowed but before stimulus information is integrated.

4/11/2011: Kacey Ballard of Brian Knutson’s lab

Multiple neural systems in reinforcement learning:

the role of mesolimbic and MTL-prefrontal circuits

Learning to take actions that lead to the most rewarding outcomes is a vital task for all organisms.  Although the mesolimbic dopamine system, with its elegant correspondence to reinforcement learning algorithms, is widely implicated in solving value-based learning problems, other neural systems are equally important for learning, particularly the hippocampus, related MTL structures, and the lateral prefrontal cortex, which are critical for encoding explicit, declarative knowledge and maintaining information in working memory over delays.  Thus, to investigate the interaction among these neural systems, we implemented an instrumental learning task with delayed feedback while subjects underwent FMRI scanning.  Using a standard model of reinforcement learning fit to subjects’ behavior, we isolated brain regions involved in calculating prediction errors and maintaining value representations.  We included a secondary distractor task that occupies attention and working memory to further test reliance on declarative, explicit strategies in what is typically thought to be a habitual, implicit form of learning.  Results showed that both mesolimbic (NAcc and MPFC) regions and MTL-prefrontal regions are recruited to support prediction error and value signals, but MTL regions are specifically impaired by the distracting secondary task, while mesolimbic regions are not.  Additionally, we found that recruitment of mesolimbic and MTL-prefrontal brain regions corresponds to individual differences in learning performance. These results provide the first evidence for MTL-mesolimbic coupling during a reward learning task, and have implications for combining multiple learning and memory systems in models of reinforcement learning.

3/14/2011: Benoit Cottereau of Tony Norcia’s lab

Deep in thought: perceptual decisions about binocular disparity

    Here we use high-density EEG to determine whether early visual areas only encode the physical stimulus or also contain representations of the expected sensory input during a decision task about binocular disparity.

    In a dynamic random dot stereogram, subjects viewed a 5 deg disparity-defined disk that moved repetitively at 1 Hz, forward and back 6 arcmin from an annular surround (12° diameter) presented in the fixation plane. Intermittently (30% of time), the disparity pattern was increased. The subjects were asked to detect these changes. We compared responses from trials where changes were correctly identified (hits) to responses from trials in which no change occurred, again correctly identified (correct rejects) within several fMRI-defined ROI’s in visual cortex. In V3A and LOC, the early response, consisting of a peak at about 200ms, reflected the magnitude of the disparity modulation. Later components of the response to the detected intermittent changes in modulation, i.e. the hits, were very different to the correct rejects. They were also different from those observed at the same time in the trials where the change wasn’t detected (misses). By time-locking the response analysis to the button press, we determined that these later components occurred at a fixed time before the subject’s response and were therefore predictive of the time of the response. Response-locked activity was also found in frontal cortex. This activity began before the button-locked response in occipital cortex.

    To test if these signals were associated to the visual inputs, we recorded the EEG responses to an omitted step. In this case, the disparity-defined disk failed to appear 30% of the time. The analysis of the hit trials showed that the earliest part of the V3A and LOC responses disappeared, but that the late part was still present and always preceded the button press. This late activity is thus due to feedback.

    Altogether, our results suggest that the V3A and LOC ROIs may encode a mismatch between the expected pattern and the actual sensory input, permitting the subject to correctly identify targets and that activity in these areas is causally related to the timing of the decision as reflected by the button press latency.


2/14/2011: Bokyung Kim of Sam McClure’s lab

The magnitude effect in intertemporal choice results from increased self-control

     People value options less as a monotonic function of their delay until delivery and choose immediate rewards over larger, delayed rewards due to this tendency. Previous work on intertemporal choices has shown that discount rates decrease dramatically as the value of individual options is increased, a phenomenon known as the magnitude effect. The dominant explanation for this finding depends on curvature of utility functions as at larger magnitudes. We report that this explanation is insufficient. In three experiments we demonstrate that the magnitude effect remains under the mere impression of larger magnitudes as created by changing the denomination of monetary offers. For example, Koreans in America and Americans in Korea are significantly more patient for Korean Won than US Dollars (approx. 1100:1 conversion rate). We hypothesized that the magnitude effect results from enhanced self-control, rather than systematic differences between subjective values. We find support for this hypothesis in a functional MRI experiment with 15 subjects instructed to choose between smaller/immediate rewards and larger/delayed rewards. We analyze brain activity in two conditions – high and low magnitude. We find that higher magnitude recruits more activation in the lateral prefrontal cortex and posterior parietal cortex. Previous studies have shown that these areas are involved in executive functions including self-control. Additionally, there was no difference in dopamine-related regions across conditions. This study has confirmed that deciding in favor of long-term outcomes, especially for higher stakes, depends on the exertion of cognitive control and this effect can be evoked by the mere reframing of outcomes. 

1/10/2011: Valerio Mante of Bill Newsome’s lab

Context-dependent gating of sensory signals for decision making

    Humans and animals process sensory signals in a remarkably flexible manner. Depending on the context in which they are perceived, identical sensory stimuli can lead to different motor actions. To study the neural processes underlying such context-dependent behavior, we trained a macaque monkey to perform two different perceptual discriminations on the same set of visual stimuli. On separate trials the monkey was instructed to either discriminate the direction of motion or the color of a random-dot display, and to report his choice with a saccade to one of two visual targets. While the monkey performed this task we recorded extracellular responses simultaneously in two cortical areas: area MT, which is thought to represent the sensory evidence relevant for direction but not color discriminations; and the frontal eye fields (FEF), which are thought to integrate the sensory evidence in favor of a particular choice. We find that MT choice probabilities are significantly larger than chance only for direction discrimination, suggesting that MT contributes to the monkey's choices during direction, but not color discrimination. Average MT responses are virtually identical across contexts, implying that gating of sensory responses does not require the modulation of firing rates in the relevant sensory areas. FEF responses typically reflect the upcoming choice in both contexts. However, differences in choice-predictive responses during direction and color discrimination suggest that the mapping between sensory evidence and FEF responses is not easily explained by standard integration-to-bound models.

12/13/2010: Winter break - No MAD lunch

11/8/2010: Alan Gordon of Anthony Wagner's lab

Neural Representations of Evidence In Mnemonic and Perceptual Decisions.

10/11/2010: Brian Knutson and Charlene Wu

Neural substrates of financial expectations

    Financial theory relies on statistical parameters (e.g., expected value, variance) to set policies and predict changes in the market. While market changes presumably depend upon human reactions to financial statistics, the neural and psychological impact of statistics on financial choice is not well understood.  Brian Knutson will describe a program of research examining individuals' neural and affective reactions to expected value and risk (the first and second statistical moments), and Charlene Wu will describe an extension of this research to skewness (the third statistical moment). Together, we conclude that our understanding of individuals' reactions to financial statistics and subsequent decisions can be improved by considering the potentially mediating role of anticipatory affect. 

9/13/2010: Jay McClelland

Decision Dynamics and Decision States in the Leaky Competing Accumulator Model

    Classical theory starting with Signal Detection Theory and the Sequential Probability Ratio Test has provided a foundation for considering the optimal policy for decision making as observers accumulate uncertain information. Certain features of human data run counter to the predictions of the classical models in their purest forms and have motivated exploration of mechanistic computational formulations that can be related both to issues of optimality and to principles of neurodynamics.

    I will describe ongoing work of this kind in collaboration with Juan Gao, Marius Usher, and others.  The focus will be on new behavioral data and the leaky competing accumulator model of the accumulation of uncertain stimulus information under externally-imposed limits in the time allowed to reach a decision.  We are exploring a version of the model in which mutual inhibition between accumulators corresponding to choice alternatives outweighs the leakage or decay of activation.  This version of the model provides a way of understanding an extensive body of recent data, and introduces a hybrid type of decision state, which has features of continuous as well as discrete decision state representations.

8/16/2010: Daniel Kimmel from Bill Newsome's lab

Neural correlates of value-based decisions in the monkey

    For many decisions, we must explicitly compare the value of two or more goods being offered. However, often decisions are not between multiple goods, but rather between a single offer and the choice to pass on that offer, such as when deciding to buy a new car, marry a significant other, or come to this talk! For these decisions the relevant comparison is between the expected benefit of the offer and its associated cost. We studied these cost-benefit decisions in the macaque monkey while recording from single neurons in the orbitofrontal cortex (OFC), which has been implicated previously in decisions between two competing goods. We found that the animal was sensitive to the balance of cost and benefit. That is, his willingness to accept an offer increased monotonically as we increased the benefit. On average, we found that neurons in the OFC encoded these valuations in a similarly graded manner, with the neural response to the offer increasing parametrically with increasing benefit. An open question is the exact role of these neurons in cost-benefit decisions. For example, do neurons encode only the expected benefit of an option, or also the cost involved in obtaining it? Are these responses merely correlated with value, or do they play a causal role in guiding value-based choice?

7/12/2010: Faraz Farzin from Tony Norcia's lab 

Decision-making in patients with amblyopia

    Perceptual decision-making involves the process of gathering and representing sensory evidence in order to guide our actions in the world, and we are constantly faced with this challenging task. The aim of the current study was to examine decision-making abilities in patients with strabismic amblyopia, a developmental disorder of visual processing which often leads to cortical suppression of input from the amblyopic eye. The talk will include behavioral data obtained using a variant of the Eriksen flanker task, in which we tested decision-making with and without the presence of stimulus and response competition. These data show that patients with amblyopia show significant delays in decision-related reaction time, particularly in the presence of conflict, and that these delays are not explained by differences in simple reaction time. Our results suggest that amblyopia has consequences not only for the development of visual function, but that it also impacts the development of higher-level decision-making networks.

6/14/2010: Kickoff meeting

    Short presentations by the labs of Anthony Wagner, Bill Newsome, Jay McClelland, Brian Knutson, Sam McClure, Tony Norcia.

Upcoming talks:

Brian Knutson and Josiah Leong

Friday, May 22nd noon-1:30p

Jordan Hall, room 102

*NOTE new day*

A BIG thanks to our sponsors:

Prof Brian Knutson and BioX