Fifteenth Annual Summer Interdisciplinary Conference

Authors, Titles, Abstracts

Listing by speaker

SpeakerAnders, Royce
Author 1Anders, Royce
Aix-Marseille University, France
Author 2Oravecz, Zita
Penn State University, USA
Author 3Alario, F.-Xavier
Aix-Marseille University, France
TitleNew Developments in Reaction Time Analysis: Merging Regression with Sequential Information Sampling Models
AbstractThis talk will focus on new developments for analyzing reaction time (RT) data at greater depths. Most notably, I will discuss a new framework that we have developed for bridging together two of the major approaches for RT analysis: regression and sequential sampling (or evidence accumulation) models. While these approaches have been previously used separately to analyze RTs differently, we have developed a new framework which allows the benefits of these approaches to be combined. Specifically in contrast to traditional evidence accumulation, the combined approach allows for more experimental factors to be analyzed jointly (by economizing the number of observations needed per experimental design cell); an explicit/simultaneous modeling of specified experimental main effects, interactions, covariates, and inter-parameter correlations; and provides the capacity to fit or predict incomplete data (e.g., missing/unobserved experimental design cells). These valuable advantages previously associated with just regression, are hence combined with the tradition of evidence accumulation, which brings its own notable contributions to better understand the cognitive mechanisms underlying the RT data. I will also share other notable developments for detailed RT distribution analysis, coupled with cognitive process modeling.

SpeakerAverbeck, Bruno
Author 1Averbeck, Bruno
TitleMarkov decision processes and the explore-exploit trade-off
AbstractIf a new restaurant opens in your neighborhood do you stop in for dinner, or do you stick with one of your familiar options that you know has reliably good food? More generally, should you exploit choice options of known reward value, or should you explore novel options that may turn out to be more rewarding? This question is known as the explore-exploit trade-off, and it is fundamental to learning in dynamic environments. In reinforcement learning, this trade-off is often managed by manipulating the amount of noise in decision processes -- sometimes choosing options that are not as rewarding or are not known to be rewarding, to learn about them. It is possible, however, to explore in a directed way if you have some knowledge of the statistics of your environment. Directed exploration is information gathering. I will discuss a theoretical framework that allows one to carry out optimal directed exploration. The framework defines the conditions under which exploration is more or less valuable. We will also examine some situations where we can use the framework to characterize behavior.

SpeakerBhatia, Sudeep
Author 1Bhatia, Sudeep
University of Pennsylvania
TitleModelling Associative Judgment
AbstractI study associative processing in high-level judgment using vector space semantic models. I find that semantic relatedness, as quantified by these models, is able to provide a good measure of the associations involved in judgment, and in turn predict responses in a large number of existing and novel judgment tasks. My results shed light on the representations underlying judgment, and highlight the close relationship between these representations and those at play in language and in the assessment of word meaning. In doing so, they show how one of the best-known and most studied theories in decision making research can be formalized in order to make precise quantitative a priori predictions for a large class of natural language judgment problems.

SpeakerBrown, Gordon
Author 1Brown, Gordon
University of Warwick
Author 2Walasek, Lukasz
University of Warwick
TitleThe Stability of “Unstable” Preferences and Judgments: A Rank-based Approach to Choice
AbstractModels in economics and psychology typically aim to identify invariances in people's behaviour — i.e., descriptions of choices that remain true across different contexts. It is typically assumed that people can be understood as possessing internal stable preferences (e.g., for cheese of a given sharpness, or beer of a particular strength), and that behavioural invariances can be expressed in terms of such preferences. However, the pervasive context-dependence of judgement and choice suggests that either (a) choices cannot be understood in terms of preferences, or (b) preferences are unstable and hence that there are no invariances to be identified. Moreover, the conventional approach assumes that people can compare different attribute values on a common scale, yet people often cannot do this (the incommensurability problem). Here we offer an alternative approach to identifying invariances: Relative Rank Theory (RRT). RRT abandons both the idea that everyday choices result from options being valued on a single scale, and the idea that invariances in choice can be understood in terms of underlying preferences expressed in real-world coordinates. Instead, RRT assumes that utility values are not available and that choices are based on judgements of relative rank constructed from binary ordinal comparisons. Thus invariances in choice may require preferences to be stated in rank-based, rather than real-world, coordinates. Different attributes may only be compared when different dimensions may be mapped into a common third dimension by a process called relative rank matching.

SpeakerCheng, Patricia
Author 1Cheng, Patricia
TitleHow causal invariance as an aspiration shapes our causal representation of the world
AbstractCausal invariance is the sameness of how a cause operates across contexts. This talk will explain and illustrate the three roles of causal invariance — aspiration, criterion for hypothesis revision, and default assumption — in scientific reasoning and intuitive causal reasoning, the latter in humans and in rats. The aspiration role drives the other two. These roles are essential to constructing a stable causal representation that is useable, that is, generalizable from the learning context to an application context.

SpeakerCowell, Rosemary
Author 1Cowell, Rosemary
University of Massachusetts Amherst
Author 2Sadil, Patrick
University of Massachusetts Amherst
TitleA Computational Model of Perceptual Deficits in Medial Temporal Lobe Amnesia
AbstractDamage to the Medial Temporal Lobe (MTL) has long been known to impair declarative memory. More recent evidence suggests that MTL damage also impairs visual perception. One recent study of visual discrimination behavior revealed a surprising anti-perceptual learning effect in MTL-damaged individuals: with exposure to a set of visual stimuli, discrimination performance worsened rather than improved (Barense et al., 2012). We present a computational model that explains this paradox by assuming that difficult visual discriminations are performed using a familiarity heuristic (i.e., subjects compare the relative familiarity of the two to-be-discriminated items). We simulate these results by instantiating the familiarity heuristic mechanism for visual discrimination in a neural network model that was previously used to simulate impairments in recognition memory (Cowell et al., 2006). The model thus accounts for both mnemonic and perceptual deficits caused by MTL damage using a unified architecture and mechanism.

SpeakerCriss, Amy
Author 1Criss, Amy
Syracuse University
Author 2Kilic, Asli
Author 3Malmberg, Ken
University of South Florida
Author 4Fontaine, Jessica
Syracuse University
TitleFeedback and Interference in Memory
AbstractUnderstanding interference in episodic memory is critical. Studies have shown a small and somewhat unreliable decrease in accuracy when items are added to the study list, called the list length effect (LLE) and a rather large and robust decrease in accuracy when items are added to a test list, termed output interference (OI). We simultaneously evaluate the effects of adding items during study and adding items during test. Feedback presented during test played an unexpected role. When feedback was present, OI was smaller in magnitude and the size of the LLE was larger. Here we present efforts to model the role of feedback within the REM model.

SpeakerDonkin, Chris
Author 1Donkin, Chris
UNSW Australia
Author 2Nosofsky, Robert
Indiana University
TitleExploring Bayesian decision rules in visual working memory
AbstractThe most successful model of visual working memory posits that visual working memory resources are distributed among to-be-remembered items in a continuous and varied manner. The model requires no explicit guessing process, and yet is capable of capturing the many documented cases of guessing - responses that are independent of the memoranda. However, the apparent guessing behavior is driven by the Bayesian decision-making machinery, and the (arguably implausible) knowledge to which it is granted access. Consideration of this model of visual working memory reveals to two important issues with respect to Bayesian models of cognition. First, the psychological plausability of the knowledge or information granted to the model is of critical importance, though difficult to discern. Second, the discarding of unreasonable 'ideal-observer' assumptions will often yield infinitely many possible Bayesian models, making it difficult to argue that such models are necessarily optimal or normative.

SpeakerDunn, John
Author 1Dunn, John
University of Adelaide
Author 2Stephens, Rachel
University of New South Wales
Author 3Hayes, Brett
University of New South Wales
Author 4Anderson, Laura
Binghamton University
TitleTesting models of reasoning with signed difference analysis.
AbstractAn ongoing debate is whether human reasoning is best accounted for by one or two processes. Single process models propose that reasoning judgments are based on the assessment of a single kind of argument strength (such as conditional probability) while dual process models propose a distinction between an intuitive or Type 1 judgment and a deliberative or Type 2 judgment. These models are frequently examined in the context of a task in which participants are asked to judge either the logical validity of an argument or the plausibility of its conclusion given the truth of the premises. The combination of two kinds of instruction and two kinds of argument (valid/invalid) leads to four dependent variables which can be modelled using basic principles of signal detection theory. This generates 6 distinct models that vary on the number of dimensions (one vs. two) and the number of distinct response criteria (0, 1, or 2). However, these models cannot be fit directly because the nature of the distributions of argument strength is unknown (and probably not normal). For this reason, we test the models using signed difference analysis (Dunn & James, 2003). This can be viewed as a “high dimensional” state-trace analysis and, like state-trace analysis, allows predictions to be generated under arbitrary monotonic transformations of the dependent variables. We apply this approach to a database of 204 observations drawn from over 20 different studies in order to determine which, if any, model can account for these data.

SpeakerFranconeri, Steven
Author 1Franconeri, Steven
Northwestern University
TitleNow you see it: Visual exploration and communication of patterns in data
AbstractWithin a well-designed visualization, your eyes can be a powerful tool for exploring and understanding patterns in data. But within a poorly-designed depiction of the same data, the same tasks can be inefficient, or even overwhelming. We'll discuss how data visualization relies on the systems that we use to perceive the natural world, and how research into the power and limits of those systems inspires prescriptions for effective visualization design.

SpeakerFrench, Bob
Author 1French, Bob
LEAD-CNRS UMR 5022/ University of Burgundy-Franche-Comté
Author 2Addyman, Caspar
Goldsmiths, University of London
Author 3Mareschal, Denis
Birkbeck, University of London
Author 4Thomas, Elizabeth
INSERM U1093 University of Burgundy-Franche-Comté
TitleUnifying prospective and retrospective interval-time estimation: A new fading-Gaussian activation-based model of interval-timing
AbstractHass and Hermann (2012) have shown that only variance-based processes will lead to the scalar growth of error that is characteristic of human time judgments. Secondly, a major meta-review of over one hundred studies (Block et al., 2010) reveals a striking interaction between the way in which temporal judgments are queried (i.e., retrospectively or prospectively) and cognitive load on participants' judgments of interval duration. For retrospective time judgments, estimates under high cognitive load are longer than under low cognitive load. For prospective judgments, the reverse pattern holds, with increased cognitive load leading to shorter estimates. We describe GAMIT, a Gaussian spreading-activation model, in which the sampling rate of an activation trace is differentially affected by cognitive load. The model unifies prospective and retrospective time estimation, normally considered separately, by relating them to the same underlying processes, in particular, attentional resource sharing. The scalar property of time estimation arises naturally from the model dynamics and the model shows the appropriate interaction between mode of query and cognitive load.

SpeakerGorea, Andrei
Author 1Gorea, Andrei
Université Paris Descartes and CNRS
Author 2Lisi, Matteo
Université Paris Descartes
Author 3Mongillo, Gianluigi
Université Paris Descartes and CNRS
TitleConfidence levels during perceptual decision-making are discrete not continuous
AbstractAnimals (including humans) are able to assess the quality of incoming sensory information and act accordingly while taking decisions. The computations underlying such ability are unclear. If neuronal activity encodes probability distributions over sensory variables, then uncertainty – hence confidence – about their value is explicitly represented and, at least in principle, readily accessible. On the other hand, if neuronal activity encodes point-estimates, then confidence must be obtained by comparing the level of the evoked response to fixed (possibly learned) criteria. To address this issue we developed a novel task allowing the behavioral read-out of confidence on a trial-by-trial basis. Each trial consisted of two consecutive decisions on whether a given signal was above or below some reference value, call it zero. The first decision was to be made on a signal uniformly drawn from an interval centered at zero. Correct/incorrect responses resulted into signals uniformly drawn from the positive/negative sub-intervals to be judged when making the second decision and subjects were told so. The task reliably elicited confidence assessments as demonstrated by the finding that second decisions were more frequently correct than first decisions. We compared the ability of Bayesian and non-Bayesian observers to predict the empirically observed pattern of both first and second decisions. The non-Bayesian observer was designed to have discrete confidence levels instantiated by one, two or three second-decision criteria representing different levels of the evoked response. Different confidence levels resulted into different level depended on the amplitude second-decision criteria. Synthetic data-sets reliably discriminated Bayesian from non-Bayesian observers. The non-Bayesian observer with two-three confidence levels systematically (over 9 subjects) outperformed the Bayesian observer in predicting the actual behavior. Hence, contrary to previous claims, confidence appears to be a discrete rather than continuous quantity. Simple heuristics are sufficient to account for confidence assessment by humans making perceptual decisions.

SpeakerGuest, Olivia
Author 1Guest, Olivia
University of Oxford
Author 2Love, Bradley
University College London
TitleWhat the Success of Brain Imaging Implies about the Neural Code
AbstractThe success of fMRI places constraints on the nature of the neural code. The fact that researchers can infer similarities between neural representations, despite limitations in what fMRI measures, implies that certain neural coding schemes are more likely than others. For fMRI to be successful given its low temporal and spatial resolution, the neural code must be temporally and spatially smooth. Through proof and simulation, we evaluate a number of reasonable coding schemes and demonstrate that only a subset are plausible given both fMRI's successes and its limitations in measuring neural activity. Coding schemes that borrow from deep belief networks and classical connectionist approaches are recoverable in simulated fMRI studies. These results suggest that brain models and neural measures need not be extremely fine grained to bridge between brain and behaviour.

SpeakerHalpern, David
Author 1Halpern, David
New York University
Author 2Gureckis, Todd
New York University
TitleConstrained Optimization of Measurements: A model of information selection in categorization
AbstractSelective attention has been a crucial feature in many theories of categorization. In many influential theories (Nosofsky 1986, Kruschke 1992), selective attention is implemented through a set of decision weights such that more relevant features of the category are weighted higher in the final categorization decision. In this project we instead take the view that attention can instead be viewed as a type of information sampling. Under this view, allocating attention to a feature means collecting more information about that feature so as to reduce uncertainty about the exact value that feature takes on. Since certain features may be more relevant to categorization and only a limited amount of information can be collected, an attention allocation is optimal when it maximally reduces uncertainty in the final categorization decision. We present a theoretical analysis, and empirical study, examining how people decide to gather more or less precise information about features prior to making a categorization decision.

SpeakerHolden, John G.
Author 1Garcia, Olivia P.
University of Cincinnati & Cintas Corp.
Author 2Holden, John G.
University of Cincinnati
TitleIntentional Control of Cognitive Dynamics
AbstractA series of lexical decision studies that manipulate speed-accuracy trade-offs are used to test the basic hypothesis that intentional change is reflected in performance in the same way that control parameters influence the dynamics of artificial complex networks.

SpeakerHolmes, William
Author 1Holmes, William
Vanderbilt University
Author 2Trueblood, Jennifer
Vanderbilt University
Author 3Heathcote, Andrew
University of Tasmania
TitleA new framework for modeling decisions about changing information
AbstractIn the real world, people integrate non-stationary information that changes systematically while the decision is in progress. Although theories of decision-making have traditionally been applied to paradigms with stationary information, non-stationary stimuli are now of increasing theoretical interest. We used perceptual stimuli that change at discrete time points during decisions along with cognitive modeling to investigate how the decision process is updated when a stimulus changes. To investigate the underlying changes in the decision process after the stimulus change, we developed a series of Piecewise Linear Ballistic Accumulator models (PLBA). The PLBA is efficient to simulate, enabling it to be fit hierarchically to participant choice and response-time distribution data using probability density approximation (PDA) methods. We similarly develop and fit a Piecewise Drift Diffusion Model (PDDM) to compare and contrast the inferences of the two models.

SpeakerHuber, David
Author 1Huber, David
University of Massachusetts, Amherst
Author 2Solstad, Trygve
Norwegian University of Science and Technology
TitleEpisodic Memory and Spatial Navigation in the Medial Temporal Lobe
AbstractHippocampal place cells and entorhinal grid cells are abundant, suggesting that the medial temporal lobe (MTL) primarily supports spatial navigation. At the same time, neuropsychological studies find that the MTL supports the formation of episodic memories. We explain this seeming contradiction with a hierarchical memory model in which episodic memories are points in a high dimensional space. Because the X/Y dimensions used to analyze place and grid cells are part of this representation, different levels of the hierarchy give the appearance of place and grid cells when only analyzed in terms of the X/Y plane. We propose that X/Y position information is provided by border cells, which is combined with the true attributes of grid cells (e.g., something other than X/Y, such as temperature, surface texture, time of day, etc.) to produce cognitive maps and ultimately multidimensional episodic memories in the hippocampus. Place cell responses are retrieved memory responses that occur when the animal is in a position sufficiently close to the location of a prior experience. Memory consolidation separates memories in the multidimensional space, producing a hexagonal array of place cells. The response of a cell representing a non-spatial attribute common to the entire set of memories is hexagonal in the X/Y plane owing to excitatory feedback from each place cell. Thus, rather building place cells out of grid cells, as proposed by other models, our account builds grid cells out of place cells. Furthermore, our account proposes that place cells are episodic memories and grid cells are non-spatial.

Speakerjones, matt
Author 1jones, matt
university of Colorado
TitleLogical incoherence of game-theoretic rationality
AbstractModern normative game theory is founded on extending instrumental rationality to cases of multiple agents: Not only do individuals choose rationally, but they expect each other to behave rationally as well. Thus rational behavior is defined recursively, and in fact circularly. I argue that this game-theoretic notion of rationality is logically inconsistent and hence meaningless. I do this first by building on analysis of the one-shot prisoner's dilemma by Shiffrin, Lee, and Zhang. I then focus on the celebrated backward induction argument for finitely repeated games, using the centipede game as an example. I prove that, following certain paths in the game's decision tree, a rational player's behavior is undefined. Consequently, the opponent cannot evaluate the options at earlier steps, and has no basis for action. Recursive rationality leads to paralysis. Time permitting, I will discuss some alternative theoretical approaches that might avoid this problem.

SpeakerKouider, Sid
Author 1Kouider, Sid
CNRS and Ecole Normale Supérieure
Author 2Goupil, Louise
CNRS and Ecole Normale Supérieure
Author 3Gelskov, Sofie
CNRS and Ecole Normale Supérieure
TitleDeveloping a reflective mind: consciousness, predictive coding and metacognition in the infant brain
AbstractThis talk will focus on whether and how infants 1) experience perceptual consciousness, 2) rely on bayesian inference during perception, and 3) rely on metacognitive sensitivity to track their own behaviors. I will first describe how one can test for perceptual consciousness in infants by relying on neural signatures of consciousness validated in adult populations. Our studies confirm the presence of these neural signatures in 5 to 15 month-old infants, albeit much slower. Regarding predictive coding, we combined EEG with a cross-modal cueing paradigm and show that neural responses for unexpected events are increased in 12 month-olds. However, this effect of prediction error was observed only during late processing stages. Early neural components, by contrast, revealed an amplification for predicted rather than surprising events, suggesting that selective attention enhances perceptual processing for expected events. These results demonstrate that the neural mechanisms underlying the use of predictive signals are already functional in infancy, but follow different dynamics depending on whether expected events are confirmed or instead surprising. Regarding metacognition, we demonstrate that infants reflect upon their own decisions to evaluate their accuracy and adapt subsequent behaviour. We show that after performing a binary choice, infants display appropriate decision confidence for correct compared to incorrect decisions. Furthermore, we show that an electrophysiological marker of error detection, the Error-Related Negativity, is elicited when 12 month-old infants make an incorrect decision. Hence, although explicit forms of metacognition might mature later during childhood, the mechanisms responsible for metacognitive sensitivity are already functional during the first year of life.

SpeakerKouider, Sid
Author 1Kouider, Sid
CNRS and Ecole Normale Supérieure
Author 2Goupil, Louise
CNRS and Ecole Normale Supérieure
Author 3Gelskov, Sofie
CNRS and Ecole Normale Supérieure
TitleDeveloping a reflective mind: consciousness, predictive coding and metacognition in the infant brain
AbstractThis talk will focus on whether and how infants 1) experience perceptual consciousness, 2) rely on predictive codes during perception, and 3) rely on metacognitive sensitivity to track their own behaviors. We will first examine to test for perceptual consciousness in infants by relying on neural signatures of consciousness validated in adult populations. Our studies confirm the presence of these neural signatures in 5 to 15 month-old infants, albeit much slower. Regarding predictive coding, we combined EEG recordings with a cross-modal cueing paradigm and show that neural responses for unexpected events are increased in 12 month-olds. However, this effect of prediction error was observed only during late processing stages associated with perceptual consciousness. Early neural components, by contrast, increased for predicted rather than surprising events, suggesting that selective attention enhances perceptual processing for expected events. These results demonstrate that predictive codes are already functional in infancy, and reveal a privileged link between neural surprise and consciousness in infants. Regarding metacognition, we demonstrate that infants reflect upon their own decisions to evaluate their accuracy and adapt subsequent behaviour. We show that after performing a binary choice, infants display appropriate decision confidence by persisting more for correct compared to incorrect decisions. Furthermore, we show that electrophysiological marker of error detection are elicited in 12 month-old infants. Hence, although explicit forms of metacognition might mature later during childhood, the mechanisms responsible for metacognitive sensitivity are already functional during the first year of life. I will conclude on perspectives for learning and education.

SpeakerLandy, David
Author 1Landy, David
Indiana University
TitleSystematic structure in measures of ignorance and political misinformation
AbstractMajor polling companies (Gallup, Pew, and Ipsos Mori) as well as academic surveys have attempted to assess people's knowledge, ignorance, and misinformation regarding politically important magnitudes, such as the proportion of the population that is foreign-born. Such surveys generally examine small sets of questions grouped by content area, and treat estimation error as indications of biased beliefs about true values. Prior research has sought explanations for errors in terms of person-specific or group-specific biases, for instance fear of minorities by majority groups (Wong, 2007) or a lack of familiarity with minority populations (Sigelman & Niemi, 2001). In contrast, we find that patterns previously reported as domain-specific bias are highly systematic across a wide range of different topics and are well characterized by traditional psychophysical models of probability estimation. After accounting for systematic estimation error, we find that biased beliefs often still exist, but that this bias is frequently in a direction opposite of what has been characterized by previous research. Furthermore the structure of this psychophysical response function is related in predictable ways to participants' levels of numeracy and political knowledge. We propose that previously reported biases must be reinterpreted; much apparent bias is best explained in terms of general cognitive factors rather than topic-specific ignorance. True misinformation is concealed by current analytic methods.

SpeakerLewandowsky, Stephan
Author 1Lewandowsky, Stephan
University of Bristol
Author 2Freeman, Mark
Loughborough University
Author 3Mann, Michael
Penn State University
TitleHarnessing the uncertainty monster: Putting quantitative constraints on the intergenerational social discount rate
AbstractThere is broad consensus among economists that unmitigated climate change will ultimately have adverse global economic consequences, that the costs of inaction will likely outweigh the cost of taking action, and that social planners should therefore put a price on carbon. However, there is considerable debate and uncertainty about the appropriate value of the social discount rate, that is the extent to which future damages should be discounted relative to mitigation costs incurred now. We briefly review the ethical issues surrounding the social discount rate and then report a simulation experiment that constrains the value of the discount rate by considering 4 sources of uncertainty and ambiguity: Scientific uncertainty about the extent of future warming, social uncertainty about future population and future economic development, political uncertainty about future mitigation trajectories, and ethical ambiguity about how much the welfare of future generations should be valued today. We compute a certainty-equivalent declining discount rate that accommodates all those sources of uncertainty and ambiguity. The forward (instantaneous) discount rate converges to a value near 0% by century's end and the spot (horizon) discount rate drops below 2% by 2100 and drops below previous estimates by 2070.

SpeakerLove, Bradley
Author 1Love, Bradley
University College London
Author 2Riefer, Peter
University College London
TitleCoherency Maximizing Exploration in the Supermarket
AbstractIn uncertain environments, effective decision makers balance exploiting options that are currently preferred against exploring alternative options that may prove superior. For example, a honeybee foraging for nectar must decide whether to continue exploiting the current patch or move to a new location. When the relative reward of options changes over time, humans explore in a normatively correct fashion, exploring more often when they are uncertain about the relative value of competing options. However, rewards in these laboratory studies were objective (e.g., monetary payoff), whereas many real-world decision environments involve subjective evaluations of reward (e.g., satisfaction with food choice). In such cases, rather than choices following preferences, preferences may follow choices with subjective reward (i.e., value) constructed to justify choice and maximize coherency. If so, the tendency to explore should lessen as uncertainty increases, contrary to previous findings. To evaluate this possibility, we examined the exploratory choices of more than 300,000 individuals in supermarkets over several years. Consumers' patterns of exploratory choice ran counter to normative models for objective rewards – the longer the exploitation streak for a product, the less likely were people to explore an alternative. These findings suggest interventions to promote healthy lifestyle choices.

SpeakerMalmberg, Kenneth
Author 1Malmberg, Kenneth
University of South Florida
TitleToward a Model of the Problem of Autobiographical Memory
AbstractDespite decades of formal development of human memory models, these models have not been extended from accounts of list learning to accounts of the everyday memory. In this talk, I propose that everyday memory poses a problem to solved, and everyday memory problems are solved in similar ways to other problems. Specifically, everyday memory depends on general knowledge and inductive inference in order to mentally reinstate temporal context associated with past experience. I will present a preliminary data set and relate theoretical constructs of human memory to the brain's Default Mode Network.

SpeakerMatzke, Dora
Author 1Matzke, Dora
University of Amsterdam
Author 2Boehm, Udo
University of Amsterdam
Author 3Marsman, Maarten
University of Amsterdam
Author 4Wagenmakers, Eric-Jan
University of Amsterdam
TitleOn the Importance of Avoiding Shortcuts in Modelling Hierarchical Data
AbstractPsychological experiments often yield data that are hierarchically structured. Popular shortcut analysis strategies that fail to properly accommodate this hierarchical structure can result in biased conclusions. To gauge the severity of these biases, we conducted a simulation study in which we generated realistic response time data for a two-group experiment. In line with well-established theoretical results, our simulations showed that Bayesian and frequentist analyses that ignore the hierarchical data structure and rely on participant means are biased towards the null hypothesis. Analyses that take a two-step approach, submitting participant-level estimates from a hierarchical model to follow-up tests are biased towards the alternative hypothesis. Only fully hierarchical analyses of multilevel data lead to correct conclusions. We discuss the relevance of our results to clinical and neuropsychological studies that rely on Bayesian hierarchical parameter estimation.

SpeakerMcLaughlin, Anne
Author 1McLaughlin, Anne
North Carolina State University
TitleModeling response to feedback in verbal rule-based categorization tasks
AbstractFeedback was manipulated after correct and incorrect answers in a series of four experiments to determine the effects of feedback type on performance, learning, and study behavior. A general benefit was found for feedback provided after incorrect responses. However, participant behaviors were changed by altering the amount of information available in feedback after correct versus incorrect answers, showing that learners are attuned to the amount of information in feedback as well as having their attention directed by corrective feedback.

SpeakerMcNamara, Timothy
Author 1McNamara, Timothy
Vanderbilt University
Author 2Chen, Xiaoli
German Centre for Neurodegenerative Diseases, Magdeburg
TitleIndividual Differences in Cue Combination During Navigation
AbstractWe examined the manner in which people integrated visual cues and self-motion cues during spatial navigation when the two cues varied in reliability. Participants walked in immersive virtual reality from a starting location to three successive waypoints and then attempted to return to the first waypoint using (a) visual cues alone, (b) self-motion cues alone, (c) congruent visual and self-motion cues, or (d) incongruent visual and self-motion cues. Performance was statistically optimal, or nearly so, under most conditions. A striking discovery existed in the large individual differences in the extent to which participants were able to use visual cues relative to self-motion cues. These individual differences were positively correlated with the extent to which participants relied on visual cues relative to self-motion cues in double-cue conditions (c & d). Correlations among tests of spatial ability, measures of task performance, and confidence ratings showed that participants with higher mental rotation scores performed better with self-motion cues relative to visual cues, assigned greater weight to self-motion cues than to visual cues, and were more confident using self-motion cues relative to visual cues. The etiology and implications of these individual differences will be discussed.

SpeakerMelcher, David
Author 1Melcher, David
University of Trento
Author 2Wutz, Andreas
University of Trento; MIT
Author 3Drewes, Jan
University of Trento
TitleHow temporal windows and perceptual cycles organize visual cognition
AbstractA basic idea in cognitive science is that perceptual and cognitive processes take time to complete, as measured for example by reaction times, Donders' subtraction method or ERPs. More recently, there has been converging evidence that perceptual systems also have an inherent temporal structure that is present even prior to stimulus presentation. Here, I will present recent work from my lab investigating how these temporal factors may create capacity limits in perception and working memory and how temporal windows influence our subjective interpretation of events. These studies, using behavioral measures, EEG, MEG and eyetracking, suggest a link between neural oscillations, visual perception, oculomotor planning and working memory. Overall, this work points to a critical role of the brain's time frames in organizing and aligning perception, cognition and action.

SpeakerMullett, Tim
Author 1Mullett, Tim
University of Warwick
TitleAttention and Behavioural Phenomena in Choice Models
AbstractThere are a number of phenomena that have been shown to have an almost ubiquitous presence across all decision making tasks. Many of these come from behavioural measurements, for example reaction time distributions almost always show a significant positive skew. More recently findings from eye tracking or attention measures have also shown stable characteristics. This includes the Late Onset Bias (often called the gaze cascade) and an overall bias to choose the item attended for longer. Specific forms of drift diffusion models have been extended with additional assumptions and have been shown to fit the data well. However, there are a large number of existing behavioural models which could accommodate these additional assumptions equally well, but have not been tested. Our approach has not been to model or fit every extant model. Instead, we take a more general approach: we test assumptions and properties that are common to many models to see which are necessary for a model to simultaneously explain common behavioural and attention phenomena. By using extensive cognitive modelling and parameter space partitioning we show that the stopping rule is the most significant factor. Models with a relative stopping rule (where a response is made based upon the magnitude of the difference in accumulated evidence) accurately capture these common phenomena at a range of parameter values. However, models with an absolute stopping rule are unable to do so, even after incorporating a range of additional assumptions such as feed-forward inhibition, mutual inhibition and decay.

SpeakerMusca, Serban C.
Author 1Musca, Serban C.
CRPCC lab - EA 1285, European University of Brittany at Rennes
Author 2Chemero, Anthony
Departments of Philosophy of Psychology, University of Cincinnati
TitleDoes word frequency explain free recall?
AbstractIs word frequency (WF) a key explanatory factor of free recall (FR)? Different models make contrary predictions on the relation between WF and FR performance, with some models predicting a positive and others a negative relationship. All models predict a monotonic relationship, yet the parametric study of Lohnas & Kahana (2013) found a non-monotonic relationship. This begs the question of whether it is genuinely WF that is at play or some other confounding variable. To ensure "WF effect" is that of word's frequency, one must show that a WF effect is found when controlling exhaustively for other explanatory variables. Starting from a thorough task analysis of FR that we carried out, including an analysis of the environmental and informational resources available, we propose, another explaining variable in addition to those already known as potential candidates (e.g., age of acquisition, number of phonemes). Our starting point was that the statistical structure of a language shapes the neural structure of people who learn it over many years and then use it daily, and that the overall pattern of these neural structures then determines the way participants remember words in FR tasks. We hypothesized that ‘relative expectancy', a measure of the mismatch between the statistical properties of the to-be-recalled words in an experimental list and the statistical properties of the language as a whole, is a key explanatory factor in FR. Simulation of extant data and original experimental data are presented to support the idea of the relevance of the variable we propose.

SpeakerOberauer, Klaus
Author 1Oberauer, Klaus
University of Zurich
Author 2Souza, Alessandra
University of Zurich
TitleDoes rehearsal help immediate serial recall?
AbstractThe assumption that articulatory rehearsal is beneficial for immediate serial recall of verbal materials has been virtually taken for granted. Correlational evidence suggests that cumulative rehearsal in particular is beneficial for serial recall (Tan & Ward, 2008, PBR). Yet, there is no experimental evidence supporting a beneficial causal effect of rehearsal on immediate serial recall. Simulations with a generic model of serial recall revealed that a mechanistic implementation of rehearsal as a maintenance mechanism protecting representations from decay is elusive (Lewandowsky & Oberauer, in press, PsychRev). We present two experiment that manipulate the frequency and the schedule of rehearsal, one with a simple-span and one with a complex-span task. Participants were instructed to remember a list of words in serial order and engage in cumulative rehearsal. They were instructed to rehearse overtly so their rehearsal could be monitored. The instruction increased the prevalence of cumulative rehearsal in comparison to a control condition in which participants were free to rehearse as they wished. Instructed cumulative rehearsal led to better recall of words from the beginning of the list at the expense of words at the end of the list. Nevertheless, participants did not recall more words overall in the instructed-rehearsal condition than in the control condition, showing that cumulative rehearsal does not improve performance in serial recall.

SpeakerPalmeri, Thomas
Author 1Palmeri, Thomas
Vanderbilt University
Author 2Annis, Jeffrey
Vanderbilt University
Author 3Shen, Jianhong (May)
Vanderbilt University
TitleModeling Perceptual Expertise
AbstractPerceptual experts have remarkable abilities at quickly and accurately categorizing and identify objects within their domain of expertise. We have been conducting a large-scale project testing individuals with a wide range of birding expertise on their categorization, perception, and memory with the aim of developing models that explain how and why performance varies over the expertise continuum. I will explain why we test expert birders, how we recruit these perceptual experts for online web-based experiments, and how we evaluate their expertise. I will describe experiments that test their categorization and memory. Performance of individuals on these experiments is modeled using a hierarchical Bayesian version of the Linear Ballistic Accumulator model. This modeling effort allows us to characterize parameters that vary with degree of perceptual expertise and provide a starting point for developing more fine-grained models that explain how and why certain parameters vary with expertise.

SpeakerPauli, Wolfgang M.
Author 1Pauli, Wolfgang M.
Calilfornia Institute of Technology
Author 2O'Doherty, John P
Calilfornia Institute of Technology
TitleThe human striatum represents cognitive maps during higher-order appetitive Pavlovian learning
AbstractfMRI BOLD responses in the subcortical striatum have repeatedly been found to scale with reward anticipation and prediction errors during Pavlovian learning. In other work, striatal BOLD signals during multistep decision tasks have been found to also represent model-based information about state transitions during human choice behavior. Less is known about the involvement of the striatum during model-based Pavlovian learning. Here we scanned human participants with high spatial and temporal resolution fMRI (1.8 mm, 600 ms) focused on subcortical brain areas, while they participated in a sequential Pavlovian conditioning paradigm involving an appetitive outcome (a pleasant juice). Critically, this paradigm enabled an investigation of whether striatal BOLD responses only covaried with expected probability of future reward, or whether BOLD responses also encoded information about the sequence of events leading up to reward delivery. The results of our multivoxel patterns classification analyses suggest that the striatum does indeed support a cognitive map of state transitions during high-order Pavlovian learning.

SpeakerPecher, Diane
Author 1Pecher, Diane
Erasmus University Rotterdam
Author 2Canits, Ivonne
Erasmus University Rotterdam
Author 3Zeelenberg, René
Erasmus University Rotterdam
TitleGrasp compatibility effects
AbstractResponses to pictures of graspable objects are influenced by the similarity between the response action itself and the grasping actions that could be performed with the object. In particular, when the size of the grasp required to respond is matches that of the object, responses are faster than when they mismatch. According to grounded cognition theories, action potentiation is the result of the sensory-motor simulations that constitute conceptual knowledge of objects. On this account, activation of a concept such as a hammer involves simulating actions such as a full hand grip of the handle. Alternatively, the effect could be explained by task-specific stimulus-response compatibility. On this account, participants align dimensions of the stimulus and of the response, such as size. When the dimensions are aligned, responses are faster than when they are not aligned, as in the Simon effect. In several experiments, we found that the presence of response competition and the relative rather than absolute size of stimuli could explain performance. This suggests that stimulus-response competition is a more likely explanation than sensory-motor simulations.

SpeakerPezzulo, Giovanni
Author 1Pezzulo, Giovanni
National Research Council, Rome, Italy
Author 2Barca, Laura
National Research Council, Rome, Italy
TitleTracking the dynamics of multi-attribute choices: examples from inter-temporal and effort-based decisions
AbstractMany real life decisions have more than one dimension - for example, when choosing between two travel destinations, one can consider the beauty of the places, the price of the travels, and so on. Understanding how these attributes are considered, evaluated and weighted over time is still an open research question. We tracked the hand (mouse) kinematics of participants involved in two well-known multi-attribute choice paradigms: inter-temporal choices (in which subjects select between smaller-sooner vs. larger-delayed rewards) and effort-based decisions (in which subjects select between a smaller reward that requires no effort vs. a larger reward that requires some effort to be secured). Our hope was that looking at dynamical choice patterns in these tasks - specifically, measuring subject's mouse movements towards one of the two choice buttons - might "unveil" aspects of multi-attribute decisions that are more difficult to see if one only considers (say) reaction times. Indeed, we found interesting dynamical choice patterns and significant individual differences in the way subjects integrate choice attributes. For example, in inter-temporal decisions, "farsighted" subjects seem to have a strong initial bias towards pressing the "larger-delayed" button, which they sometimes correct on-line; while "discounters" show more choice uncertainty in their mouse trajectories, contrary to the idea that they are more impulsive. These results might help understanding they way multiple attributes are considered and integrated during a choice. By the same token, these results can help understanding how subjects discount the value of an offer depending on temporal delay and physical effort.

SpeakerRamscar, Michael
Author 1Ramscar, Michael
Eberhard Karls Universität Tübingen
TitleThe mismeasurement of mind: Why neuropsychological test results exaggerate "healthy cognitive decline."
AbstractPerformance on neuropsychological tests declines with age. This is taken as evidence that cognitive capacity declines across the lifespan, providing a functional characterization of structural change in the ageing brain However, neuropsychological tests do not control for accumulated learning, and thus ignore the impact that increased knowledge and expertise can have on task performance. Here, we examine the effect of formally controlling for these factors on what is considered one of the most reliable measures of lifespan cognitive decline, Paired Associate Learning (PAL). We find that age-related changes in PAL performance are entirely consistent with the predictions of the error-driven “associative learning” models that represent the gold standard in other areas of behavioral and neuroscientific research. A modeling simulation shows how the changes seen in PAL performance across the lifespan A simulation using a standard learning model shows that lifespan changes in PAL performance are consistent with increasing knowledge, and predicts that holding age constant while varying linguistic experience will produce the effects usually interpreted as age-related decline. Consistent with this, we show that in German PAL tests, older Chinese-German bilinguals outperform age-matched native German speakers, and this advantage increases with age. These results illustrate how neuropsychological tests inflate estimates of functional decline and distort our understanding of neurological change across the lifespan.

SpeakerRegenwetter, Michel
Author 1Regenwetter, Michel
University of Illinois at Urbana-Champaign
Author 2Robinson, Maria
University of Illinois at Urbana-Champaign
TitleThe construct-behavior gap in behavioral decision research: A challenge beyond replicability
AbstractBehavioral decision research compares theoretical constructs such as preferences to behavior such as observed choices. Three common methods for connecting constructs to behavior are 1) to count the total number of choices of a certain kind across participants and decision problems, 2) to compare what most people choose in each decision problem against a predicted pairwise preference, or, 3) to enumerate the decision problems in which two experimental conditions generate a one- sided significant difference in choice frequency. While simple, these methods are heuristics. They are subject to well-known reasoning fallacies, most notably the fallacy of sweeping generalization and the fallacy of composition. No amount of replication can alleviate these fallacies. The remedy to these very common problems lies in spelling out precise theories of how hypothetical constructs translate into behavior, not in successful replication of hard to interpret effects.

SpeakerRehder, Bob
Author 1Rehder, Bob
New York University
TitleBeyond Markov: The Beta-Q Model of Causal Reasoning
AbstractAlthough many theories of causal cognition are based on causal graphical models, a defining property of such models—the independence relations stipulated by the Markov condition—is routinely violated by human reasoners. Three accounts of why people violate independence are formalized and subjected to experimental test. Subjects' inferences were consistent with a model that stipulates that humans interpret causal networks as implying patterns of co-occurrence among variables that are different than those stipulated by the normative model.

SpeakerSanborn, Adam
Author 1Sanborn, Adam
University of Warwick
Author 2Tripp, James
University of Warwick
Author 3Stewart, Neil
University of Warwick
Author 4Noguchi, Takao
University College London
TitleMost people are normative some of the time: Mixtures of combination rules are used in estimates of conjunctions and disjunctions
AbstractHuman estimates of the probabilities of combinations of events show well-established violations of probability theory, most notably the conjunction and disjunction fallacies. These violations have led researchers to conclude that the rules of probability are too complex for most people to use, and that cognitively easier approximations such as averaging are used instead. However, previous work has either looked at data averaged over participants or has assumed that individuals use only a single combination rule. We collected repeated estimates of conjunctions and disjunctions and investigated whether individuals consistently used a single rule or used a repertoire of rules using a trial-by-trial Bayesian analysis. We found that most participants were best described as randomly selecting a combination rule on each trial, and that a large majority of participants use the correct rule at least some of the time.

SpeakerShiffrin, Richard
Author 1Shiffrin, Richard
Indiana University
Author 2Nosofsky, Robert
Indiana University
Author 3Cao, Rui
Indiana University
TitleLearning in short-term memory scanning
AbstractShort-term memory search does not occur in isolation but involves retrieval from long-term memory, not only due to encoding the test stimulus, but also due to learning that can occur over just a few trials. We investigate the way learning depends on consistency of stimulus and response mappings from one trial to the next, and what is learned: item-response mappings, or category-response mappings. We show that item-response mappings can be learned and used if consistent for just a few trials. Category response mappings can be learned and used but only after a great deal of learning over many trials. Inconsistency prevents learning if mappings change from trial to trial, but not if changes occur every ten trials. We demonstrate how 'switch costs' harm performance.

SpeakerSimen, Patrick
Author 1Simen, Patrick
Oberlin College, Department of Neuroscience
TitleRobust time scale invariance in timing and perceptual decision making
AbstractResponse times (RTs) in two-choice perceptual decision making tend to be time scale invariant: when you divide the RTs in different task conditions by their means, the resulting, normalized RT distributions tend to overlap. This also occurs in interval timing tasks, in which participants decide when to make a pre-ordained response after a cue. Differences in average RTs across conditions in such tasks can span orders of magnitude, but time scale invariance appears robust here. Typically parameterized random walk models of decision making, however, can only approximate time-scale invariance across decision making task conditions. Classic models of interval timing fail more spectacularly to account naturally for observed, behavioral invariances. Yet robust timescale invariance is produced by accumulator models of timing when they are generalized in precisely the same way as the well-known diffusion decision model has been generalized, in order to explain unequal correct and error RT means in perceptual decision making (the restricted model, in contrast, predicts equality when there is no response bias). I will review the parallels between the models and the behavioral phenomena of these two fields of research, and show how laws of behavior in both domains are explained by some simple parameterizations of classic diffusion models. I will also address a potentially central role of reward in parameterizing these models, and some behavioral evidence that supports a role for reward rate in adaptive behavior in decision and timing tasks.

SpeakerSloutsky, Vladimir
Author 1Sloutsky, Vladimir
Ohio State University
TitleThe Development of Categorization: Evidence from Category Learning and Recognition Memory
AbstractWhat is the mechanism of categorization and how does it change with development? To answer these questions, we conducted category learning experiments with 4- and 6-year-olds and adults. In all experiments, participants learned categories and then were tested with categorization and recognition tasks. In Experiments 1 participants ably learned the categories, and representational differences transpired between children and adults: adults exhibited better memory for most predictive features, whereas young children remembered well all the features. In Experiment 2, participants' attention was explicitly directed to the most predictive feature, and in Experiment 3, their attention was directed to the less predictive features. Adults' remembered features differentially, according to instructions, whereas young children remembered all the features equally well. Furthermore, their memory for features that were not cued was better than memory for these features in adults. Results of computational modeling indicate that whereas categorization decision in adults are predicted by their representations of the category, in children categorization decision and representations were independent. These results suggest important developmental differences in attention and representation of categories.

SpeakerSpeekenbrink, Maarten
Author 1Speekenbrink, Maarten
University College London
Author 2Schulz, Eric
University College London
TitleDecisions in context: Exploration and exploitation in contextual multi-armed bandits
AbstractContextual multi-armed bandits (CMABs) are a general framework to study how people make decisions from experience. In a CMAB, an agent is faced with a number of options (arms of the bandit) to choose from. At each time, there is a stochastic reward associated to each arm, which depends (partly) on a set of observable features (the context). The task is to maximise the reward obtained over repeated choices by learning about the reward distributions and how they relate to the contextual features. CMABs involve an interesting exploration-exploitation trade-off, as it is good to sometimes choose an option not because it is expected to give a good reward, but because it is highly informative about the relation between context and rewards. I will describe our recent theoretical and empirical work on how people solve CMAB tasks, using Gaussian Processes to represent the current knowledge about the context-rewards relations, and a variety of acquisition functions which determine how this knowledge is used to make decisions.

SpeakerSperling, George
Author 1Sun, Peng
New York UNiversity
Author 2Chubb, Charles
University of California, Irvine
Author 3Wright, Charles
University of California Irvine
Author 4Sperling, George
University of California, Irvine
TitleAttention filters for features. New results.
AbstractThis abstract is a place holder for a presentation of our most recent findings on feature attention. An attention filter is a brain process, initiated by a participant in the context of a task requiring feature-based attention, that operates broadly across space to modulate the relative effectiveness with which different features in the retinal input influence performance. The method for quantitatively measuring attention filters uses a ``statistical summary representation" (SSR) task in which the participant strives to mouse-click the centroid of a briefly flashed cloud composed of items of different types (e.g., dots of different luminances or sizes), weighting some types of items more strongly than others. In different attention conditions, the target weights for different item-types in the centroid task are varied. The actual weights exerted on the participant's responses by different item-types in any given attention condition are derived by simple linear regression. Because, on each trial, the centroid paradigm obtains information about the relative effectiveness of all the features in the display, both target and distractor features, and because the participant's response is a continuous variable in each of two dimensions (versus a simple binary choice as in most previous paradigms), it is remarkably powerful. We describe (1) algebraic derivations for three useful statistics to describe attention filters: efficiency, fidelity, and data driveness, (2) confidence bounds on these statistics, and (3) some important procedural improvements: singleton trials, constant dispersion. Illustrative examples will be shown as time permits.

SpeakerSteil, Jochen
Author 1Steil, Jochen
Bielefeld University
TitleScaling exploratory robot learning from motor to social learning
AbstractThere is a long-standing and ongoing discussion on the role of forward and inverse models in motor control and beyond. Bridging robotics and cognitive science, the talk discusses the relevance and implications of recent advancements in exploratory robot learning, summarized under the term Goal Babbling, for social and cognitive learning. Whereas Goal babbling is motivated by the early infants' capability to reach for goals, a respective computational model to explore task spaces directly without previous motor babbling has been very successfully deployed in robotics, as will be shown by examples. Motivated from this development, the talk subscribes to the notion that learning models of others is primarily a task of acquisition of an inverse model. While this is difficult to learn in general, the talk speculates that a suitable transfer of Goal Babbling ideas provides a new approach towards learning theory-of-mind and self-other distinctions in a radically interaction-centred way.

SpeakerStoianov, Ivilin Peev
Author 1Stoianov, Ivilin Peev
Aix-Marseile University and CNRS
Author 2Ziegler, Johannes
Aix-Marseile University and CNRS
Author 3Grainger, Jonathan
Aix-Marseile University and CNRS
TitleThe perceptual code in reading: explaining the spatial variance and invariance with a neurocomputational model of visual word perception
AbstractUsing a neurocomputational approach, we investigated the mechanism of visual word perception and more specifically, the highly disputable letter-position coding scheme that still lacks an adequate neural-level account. We combined a deep generative network to learn the visual perceptual hierarchy in reading and a linear map to extract the abstract letter identities from the emergent perceptual code and infer then word identity. To account for the ample availability of perceptual data and the limited evidence for word identity, we trained the generative network with unsupervised learning on a large dataset of images of words with variable features and locations and then trained the abstract letter detectors on a much smaller dataset. Surprisingly, the generative model developed hierarchical detectors of letter shapes at multiple retinal locations but no detectors of chunks of letters. Despite of that, the visual features in the deepest layer provided an adequate support for location-invariant extraction of letter- and word-identity. Moreover, imperfections in the perceptual processing provided plausible neural-level account of various phenomena characterizing position coding, including the so-called “letter-transposition effect” in the perceptual matching task. Altogether, our model provides a plausible computational hypothesis for the mechanism of visual word perception that could help to explain multiple phenomena in reading at the neural level.

SpeakerTalmi, Deborah
Author 1Talmi, Deborah
University of Manchester
TitleModelling the list composition effect on emotional memory
AbstractFree recall of emotional pictures is better than free recall of neutral pictures, but only when these two types of pictures are presented in mixed lists. Free recall of emotional pictures is equivalent to that of neutral pictures in pure lists, which contain either emotional or neutral pictures. This talk will explore how this effect of list composition can be accommodated within existing models of memory. Emotional pictures could be considered ‘strong' items, because they are processed for longer, and trigger deeper semantic analysis. This kind of strength is, however, different from the strength that is thought to underlie the list-strength effect; and a number of findings with word stimuli, and when the task is performed under divided attention, go against the list-strength interpretation of this effect. Therefore, accommodating this effect within SAM or REM is not trivial. Emotional pictures are more distinct relative to other items stored in memory, so the effect could stem from relative distinctiveness at retrieval. Yet the specific pattern of findings, where the effect is driven by a drop in neutral memory in mixed lists appears to go against the predictions of SIMPLE. Emotional pictures are likely associated with stronger prediction error and surprise when they are encoded, regardless of the context, so predictive coding models also cannot account for the list composition effect without additional assumptions. The list-composition effect on emotional memory is robust, but is currently not explained well by existing memory models.

SpeakerTestolin, Alberto
Author 1Testolin, Alberto
University of Padova
Author 2Stoianov, Ivilin
Aix-Marseille Université
Author 3Zorzi, Marco
University of Padova
TitleLetter perception emerges in deep neural networks from unsupervised learning and recycling of natural image statistics
AbstractLetter perception is the visual front-end of reading, a key human ability and a major achievement of cultural evolution. Processing of visual symbols is thought to emerge through learning and recycling of pre-existing neuronal networks for visual object recognition. We tested this hypothesis in a large-scale computational model of letter perception based on deep neural networks. In line with neuroimaging and neurophysiological evidence, our model relies on a hierarchy of increasingly more complex internal representations, which emerge through unsupervised learning by fitting a probabilistic generative model to the sensory input. Earlier processing levels recycle domain-general visual features learned from natural image patches, while domain-specific feature detectors emerge in upstream neurons following exposure to letters presented as real images in a variety of fonts and styles. We show that visual primitives extracted from natural scenes can be effectively reused for letter perception, thereby supporting the hypothesis that the shape of visual symbols has been culturally selected to match the type of statistical structure found in our environment. Crucially, abstract representations emerging from deep unsupervised learning can be easily mapped to corresponding letter identities by linear read-out, which supports robust letter recognition in high-noise conditions and produces accurate simulations of psychophysical data.

SpeakerTrcek, Denis
Author 1Trcek, Denis
University of Ljubljana
AbstractResearch on formal models for trust (and computational trust management in general) is close to its deployment in global digital ecosystems. But this kind of research, which is going on now for over twenty years, has produced a plethora of models so far. Therefore the scientific community is facing the following key questions: How can one compare these models, what metrics can be used, and how they can be effectively evaluated in terms of better or worse performance? Certain testbeds have been developed in the community to solve these questions, but with notable shortcomings. They typically evaluate trust models by combining them with some ad hoc decision making mechanism and then evaluate the quality of trust-based decisions. They assume that if using the same decision making mechanism then this very mechanism becomes irrelevant for such evaluation. We claim, however, that the choice of decision making mechanism is very relevant. To test our claim we have built an open source test-bed, called Alpha Agent testbed that can evaluate trust models either with or without decision making mechanism and rank them accordingly. But this is still just the basis for the core question – assuming a quite good understanding of what trust is and its relationship to decision making, are we able to attribute it as specifically as possible to certain areas of a nervous system?

SpeakerTrueblood, Jennifer
Author 1Trueblood, Jennifer
Vanderbilt University
Author 2Yearsley, James
Vanderbilt University
Author 3Pothos, Emmanuel
City University London
TitleWhen are causal representations quantum versus classical?
AbstractDecades of research has shown that human decision-making often violates the rules of classical probability theory. Quantum probability theory provides an exciting new framework to model human behavior. In this talk, I will compare quantum and classical probability models of human causal reasoning. We adapted a paradigm from Rehder and Hastie (2001) where participants made judgments about a simple causal scenario involving novel categories. They also completed the Cognitive Reflection Task (CRT), a simple measure of cognitive ability that distinguishes between effortful, reflective processes and those that are executed more quickly with little conscious deliberation. A Bayesian analysis revealed that participants who tended to engage quick, intuitive cognitive processes, as measured by the CRT, were better described by a quantum model. Participants that tended to engage effortful, reflective processes were better described by a classical model. Further, participants' judgments changed through the course of the experiment as they gained more experience with the task. By the end of the experiment, all participants were better described by a classical model. This suggests that learning and experience help people form a classical representation of information.

SpeakerUsher, Marius
Author 1Usher, Marius
Tel-Aviv University
Author 2Tsestos, Konstantinos
Birkbeck College, University of London
Author 3Summerfield, Chris
University of Oxford
Author 4Chater, Nick
University of Warick
Author 5Moran, Rani
Tel-Aviv University
TitleSelective-integration: a decision mechanism that accounts for violations of transitivy and of the independence axiom (decoy effects)
AbstractTransitivity and the independence on irrelevant alternatives are two of the central principles of rational decision making. While people are thought to be adaptive in their perceptual decisions, research in multi-attribute decision-making has revealed both types of violations. Here we show that a selective-integration model, which assumes that values are integrated subject to ranks, can account for both findings. Moreover, the model provides an account for why such a "selective-integration" may be in place: to protect the decision mechanism from late noise.

Speakervan Ravenzwaaij, Don
Author 1van Ravenzwaaij, Don
University of Groningen
Author 2Donkin, Chris
University of New South Wales
Author 3Vandekerckhove, Joachim
University of California Irvine
TitleThe EZ Diffusion Model Provides a Powerful Test of Simple Empirical Effects
AbstractOver the last four decades, sequential accumulation models for choice response times have spread through cognitive psychology like wildfire. The most popular style of accumulator model is the diffusion model (Ratcliff, 1978), which has been shown to account for data from a wide range of paradigms, including perceptual discrimination, letter identification, lexical decision, recognition memory, and signal detection. Since its original inception, the model has become increasingly complex in order to account for subtle, but reliable, data patterns. The additional complexity of the diffusion model renders it a tool that is only for experts. In response, Wagenmakers, van der Maas, and Grasman (2007) proposed that researchers could use a more basic version of the diffusion model, the EZ diffusion. Here, we simulate experimental effects on data generated from the full diffusion model and compare the full diffusion model and EZ diffusion on their power. We show that the EZ diffusion model, by virtue of its relative simplicity, is better able to detect experimental effects than the data–generating full diffusion model.

SpeakerWalasek, Lukasz
Author 1Walasek, Lukasz
University of Warwick
Author 2Brown, Gordon
University of Warwick
TitleIncome Inequality and Positional Consumption Online
AbstractHow does income inequality affect our concern with social status? According to a social rank hypothesis, greater concern with “keeping up with the Joneses” may be a rational response to higher income inequality. Consequently, consumers who live in regions with higher income inequality will show greater concern with, and attention towards, positional goods and high-status brands that serve a social signalling role. We tested this account using Google Correlate and Google Trends to find internet search terms that correlate (both positively and negatively) with income inequality within the US and cross-nationally. Findings were consistent with the social rank hypothesis, showing that in more unequal regions people devote more of their resources (here, time searching the web) researching high status goods such as expensive watches or luxury perfumes. We replicated these results examining Twitter, where we find that online posts about (e.g.) Prada, Gucci or Rolex are more prevalent in more unequal US states. Using sentiment analysis, we show that the positivity with which high and low status brands are mentioned on Twitter does not differ as a function of income inequality. Finally, we demonstrate that the within-nation relationship between inequality and positional consumption is unlikely to be driven by the spending tendencies of the wealthiest members of a society. We conclude with a proposal for a psychological model that links individual perception of income and wealth distribution to a range of behaviours that are critical for maintaining well-being of a person and a society as a whole.

SpeakerYearsley, James
Author 1Yearsley, James
Vanderbilt University
Author 2Trueblood, Jennifer
Vanderbilt University
TitleQuantum and Classical Models of Causal Reasoning for Political Judgments
AbstractReasoning about the causal relationships between events is an important component of cognition, allowing us to make sense of the world. Arguably, the most successful models of causal reasoning, Causal Graphical Models (CGMs), perform well in some situations, but there is considerable variation in how well they are able to account for data, both across scenarios and between individuals. Phenomena such as order effects in predictive judgments and conjunction fallacies in judgments about causes and effects cannot be accounted for by CGMs. We propose a model of causal reasoning based on quantum probability (QP) theory that accounts for behavior in situations where CGMs fail. Whether QP or classical models are appropriate depends on the representation of events constructed by the reasoner. We describe a large (N=1200) experiment conducted during the US Presidential primaries involving judgments about the outcomes of primaries and the eventual nominations. Robust order effects and conjunction fallacies were observed, but there was considerable variation in behavior across participants and candidates. This suggests that the representation of events used to reason about the world can vary between people and from task to task.

SpeakerZiegler, Johannes
Author 1Ziegler, Johannes
Aix-Marseile University and CNRS
Author 2Perry, Conrad
Swinburne University of Technology
Author 3Zorzi, Marco
Università di Padova
TitlePredicting Individual Dyslexia Patterns And Intervention Strategies Through Computational Modelling
AbstractLearning to read in alphabetic languages relies on two core mechanisms: phonological decoding and self-teaching. Here, we present the first full-blown developmentally plausible computational model of reading acquisition that implements these two mechanisms. It was used to simulate developmental trajectories of 622 children (388 dyslexics). We show that individual reading performance on words and nonwords can be simulated with high accuracy on the basis of their underlying deficits in subcomponents of the reading network. Such simulations make it possible to predict for any given child how remediating one or several subcomponents should improve reading of words and nonwords. We further show that common single-deficit theories are unable to account for the observed heterogeneity in reading performance. We thus advocate a novel multi-factorial computational approach of understanding reading and dyslexia, which has concrete practical implications for intervention.

SpeakerZorzi, Marco
Author 1Zorzi, Marco
University of Padova
TitleThe effect of attentional load on visuospatial processing in normal and damaged brains
AbstractSpatial awareness depends on the complex interplay between spatial and non-spatial cognitive processes. I will review studies from my lab that investigated the effect of multitasking on spatial monitoring using a variety of methods and subject populations (healthy individuals and patients with brain damage). Our studies show that i) spatial awareness in stroke patients can be dramatically and asymmetrically hampered by multitasking, even when patients do not show any deficit in classic neuropsychological testing; ii) a spatial awareness deficit is induced in patients with left hemisphere stroke, a population in which spatial deficits are thought to be uncommon; iii) in healthy individuals, increased attentional load during multitasking is psychophysiologically reflected by increasing pupil dilation (eye-tracking data) and early suppression of visual areas (ERP data). In summary, our multitasking approach mimics complex everyday life requirements, maximally triggers competitive mechanisms, and selectively exacerbates contralesional spatial deficits after brain damage.

Contact with questions.