Fourteenth Annual Summer Interdisciplinary Conference

Authors, Titles, Abstracts

Listing by speaker

SpeakerAverbeck, Bruno
Author 1Averbeck, Bruno
National Institutes of Health
TitleMarkov decision processes as models for choice tasks
AbstractDecision making has been studied with a wide array of tasks. In this talk I will examine and compare the theoretical structure of bandit, information sampling and foraging tasks. These tasks move beyond tasks where the choice in the current trial does not affect future expected rewards. We have modeled these using Markov decision processes (MDPs). MDPs provide a general framework for modeling tasks in which decisions affect the information on which future choices will be made. Under the assumption that agents are maximizing expected rewards, MDPs provide solutions which will maximize the number of rewards, on average. We find that all three classes of tasks pose choices among actions which trade-off immediate and future expected rewards. The tasks drive these trade-offs in unique ways, however. For bandit and information sampling tasks, increasing uncertainty or the relevant time horizon shifts value to choices that pay-off in the future. Correspondingly, decreasing uncertainty increases the relative value of choices that pay-off immediately. For foraging tasks the time-horizon plays the dominant role, as choices do not affect future uncertainty in these tasks. In addition to the modeling I will also examine behavioral performance on the same tasks. When examining behavior we will consider two questions. First, how close are subjects to the optimal models, and/or under what conditions are they more or less optimal? And, second, when subjects are suboptimal, how can the model be used to generate insight into the ways in which the subjects are not optimal.

SpeakerCheng, Patricia
Author 1Carroll, Christopher
Carnegie Mellon University
Author 2Cheng, Patricia
TitleCausal invariance and the construction of causal knowledge
AbstractCausal knowledge consists of explanations of a reality that the learner has no access to except through representations. When do people feel dissatisfied with their explanation and revise their causal hypotheses? To develop causal explanations that generalize across the learning and application contexts, one criterion is to revise an explanation when the hypothesized cause is perceived to fail to operate the same way across different situations – that is, when the assumption of causal invariance is violated. Three approaches to causal learning – the associative approach, causal Bayes nets, and the causal-power approach – treat the violation of causal invariance differently. I present an analysis of three essential aspects of the criterion of causal invariance -- as an aspiration, a defeasible default assumption, and a description -- in shaping a generalizable representation of the causal world, and illustrate differences in the treatment of a violation of causal invariance involving binary variables by the three approaches. Whereas intuitive causal learning seems to adopt causal invariance as a criterion, this criterion is adopted by neither commonly used causal inference methods in machine learning nor associative inference models in psychology or in science. Our analysis suggests that the intuitive approach has adaptive value in light of the representational nature of causal knowledge.

SpeakerCox, Gregory
Author 1Cox, Gregory
Syracuse University
Author 2Shiffrin, Richard
Indiana University
Author 3Criss, Amy
Syracuse University
TitleA Dynamic Approach to Item and Associative Recognition
AbstractAn approach to recognition that explicitly treats memory processes as dynamic has the potential to afford deeper insights into underlying mechanisms, such as those involved in criterion setting, speed-accuracy trade-off, and the nature of word frequency in episodic memory (Cox & Shiffrin, 2012; Cox & Shiffrin, in prep.). According to this approach, a recognition decision is based on a memory probe that evolves over time as features of the test item are perceived and/or retrieved from semantic memory. At any given time, the probe is compared in parallel to traces in episodic memory, which are activated to varying degrees depending on their similarity to the probe in both content (e.g., visual or semantic features) and context (e.g., time and location). The average similarity of the probe to traces in memory changes with characteristic dynamics that can distinguish between targets and foils, serving as a basis for recognition decisions. We extend this approach beyond single item recognition to the recognition of pairs and associations. Although the ability to discriminate between studied and unstudied associations has often been attributed to a recall-like process, we draw on evidence from a variety of previous studies of speed-accuracy trade-off and new studies measuring response time distributions to show that the dynamics of associative recognition are more consistent with a compound cue mechanism than with a form of recall. We present a formal model of this process that is a direct extension of the dynamic model for single-item recognition, with the key assumption that associations are represented as additional emergent features that do not join the probe until after most item features have already been accumulated. This model provides good quantitative fits to both response time distributions and speed-accuracy trade-off data in associative recognition and offers additional insight into the effects of pre-existing semantic associations among studied pairs (e.g, Dosher, 1984) and in the interpretation of ERP signatures traditionally attributed to familiarity and recollection processes (e.g., Rugg & Curran, 2007).

SpeakerCriss, Amy
Author 1Criss, Amy
Syracuse University
Author 2Aue, William
Purdue University
Title(lack of) Output interference in retrieval from semantic memory
AbstractThe benefits of testing on later memory performance are well documented, however the manner in which testing harms memory performance is less well understood. This research is concerned with the finding that accuracy decreases over the course of testing, a phenomena termed output interference (OI). OI has primarily been investigated with episodic memory, but there is limited research investigating OI in measures of semantic memory (i.e., knowledge). We present data showing no OI in tasks that require retrieval from semantic memory.

SpeakerFoster, James
Author 1Foster, James
University of Colorado Boulder
Author 2Jones, Matt
University of Colorado Boulder
TitleReliability Weighting for Relational Cue Combinations
AbstractIn previous talks, I have presented a computational model of Analogical Reinforcement Learning, in which analogy enables abstract generalization and reinforcement learning drives discovery of useful relational concepts. In this talk, I focus on an experiment with humans that supports two predictions of the model. The first prediction is that people should be able to associate reward values to categories that are defined by relational properties. This prediction applies even when no individual stimulus is ever repeated and no stimulus feature is predictive of reward. The second prediction is that people will learn how strongly to rely on different relational concepts depending on how predictive they are. The computational model learns which relational structures are more useful (i.e., more predictive of reward) and weights those structures more heavily when estimating the values of novel stimuli. In the experiment, participants were trained to select among stimuli to maximize their reward, with different relationally determined categories associated with different mean reward values. Reward variance was manipulated between categories, such that the schema defining one category was more predictive of the outcome than the schema defining the other category. Subjects learned to reliably choose stimuli from the category with the higher mean reward, indicating they could associate value with relational categories. Moreover, in generalizing reward prediction to compound stimuli (stimuli instantiating schemas from both categories), participants' reward predictions were biased toward the more predictive category. These results are consistent with both predictions of the model.

SpeakerGibson, Steven
Author 1Gibson, Steven
Northcentral University
TitleKnowledge construction: A field neutral terminology for cognition
AbstractHow do we talk about cognitive products across differing fields of research? This report summarizes a proposed terminological approach for the discussion of cognition, learning and knowledge production. The fields of cognitive research, computer science, educational research, and management studies have differing models for cognition and differing theoretical approaches. A shared approach to discussing cognitive tasks and behaviors can open new areas of research and possibly answer some outstanding questions. This shared terminology should address representation of perceptual data, memory storage and recall, attention, concept relationship and knowledge construction. This terminology should be neutral as regards theoretical approaches employed in the fields of research and be free of embedded hypotheses regarding underlying mechanisms. This report focuses on terms relating to construction of knowledge and representation of information.

SpeakerGuest, Olivia
Author 1Love, Bradley
Author 2Guest, Olivia
Author 3Kopec, Lukasz
TitleOptimism Bias in Novice and Expert Sport Forecasting
AbstractPeople are optimistic about their prospects relative to others. However, existing studies can be difficult to interpret because outcomes are not zero-sum. For example, one person avoiding cancer does not necessitate that another person develops cancer. Ideally, optimism bias would be evaluated within a closed formal system to establish with certainty the extent of the bias and the associated environmental factors, such that optimism bias is demonstrated when a population is internally inconsistent. Accordingly, we asked NFL fans to predict how many games teams they liked and disliked would win in the 2015 season. Fans, like ESPN reporters assigned to cover a team, were overly optimistic about their team's prospects. The opposite pattern was found for teams that fans disliked. Optimism may flourish because year-to-year team results are marked by auto-correlation and regression to the group mean (i.e., good teams stay good, but bad teams improve).

SpeakerHawkins, Robert
Author 1Hawkins, Robert
Stanford University
Author 2Goodman, Noah
Stanford University
TitleWhy do you ask? Good questions provoke informative answers.
AbstractWhat makes a question useful? What makes an answer appropriate? In this talk, we formulate a family of increasingly sophisticated probabilistic models of question-answer behavior within the Rational Speech Act framework. We compare these models based on three different pieces of evidence: first, we demonstrate how our answerer models capture a battery of four classic effects in psycholinguistics, which show that an answerer's level of informativeness varies with the inferred questioner goal. Second, we jointly test the questioner and answerer components of our model based on empirical evidence from a simple question-answer reasoning game. Third, we designed a real-time, multi-player version of this game with a wider range of conditions, which allows us to better distinguish among the questioner models. We find that sophisticated pragmatic reasoning is needed to account for some critical aspects of the data. People can use questions to provide cues to the answerer about their interest, and can select answers that are informative about inferred interests.

SpeakerHemmer, Pernille
Author 1Hemmer, Pernille
Rutgers University
Author 2Persaud, Kimele
Rutgers University
TitleThe Time Course of Errors in Long-Term Episodic Memory for Color
AbstractPrior knowledge is known to influence recall when episodic information is noisy. Recent approaches, however, have suggested that recall is the result of either remembering (with some noise) or guessing (e.g. Brady et al. 2013). Importantly, the error distributions are well fit by a mixture of a Gaussian-like (noisy memory) and uniform (guessing) distribution. This stands in contrast to a Bayesian assumption that recall is a combination of expectations learned from the environment with noisy memory representations. Here, we evaluate the fidelity of long-term (LT) episodic memory for color, and the contribution of imprecise recall, prior knowledge, and random guessing to memory errors. Using a continuous recall paradigm, we found that at an aggregate level, performance appears to have a high rate of guessing. However, partitioning performance by lag (i.e., the number of intervening trials between study and test) reveals changes in the distribution of error over the time course of recall. We found that, immediate LT memory mirrors perception in its high fidelity, but with increasing lag the precision of memory appears to be more complex, and at longer lags recall is a mixture of episodic information, and guessing. We speculate that performance at intermediate lags, consistent with a Bayesian assumption, reflects the influence of category knowledge on noisy episodic representations. We implement and compare several models, including a simple Bayesian memory model and the ‘remember-guess' model. Our findings suggest that, rather than the loss of fidelity in LT memory being acute, there is an intermediate stage reliant on prior knowledge.

SpeakerHendrickson, Andrew
Author 1Hendrickson, Andrew
University of Adelaide
Author 2Perfors, Amy
University of Adelaide
TitleZipfian distributions and cross-situational word learning
AbstractHow do children learn words when one word can refer to many possible referents in any given scene? One popular theory is that they can leverage the statistics of word usage across many different scenes in order to isolate specific word meanings (Yu & Smith, 2007). Both adults and children have shown impressive learning in this kind of cross-situational learning paradigm, but relatively little is still known about how well it scales to real language. Some have suggested that when words follow a Zipfian distribution (as they do in natural language), a full lexicon should not be learnable because of the many low-frequency words that are only observed a few times (Vogt 2012). In this work, we suggest the opposite: we show that when the distribution of words mimics that in natural speech -- i.e., when it is Zipfian -- adults show improved learning in cross-situational contexts. Over a series of experiments, we show further that this effect extends beyond the high-frequency words: when matched for word frequency, Zipfian distributions produce better learning than a uniform distribution. Implications for theories of word learning, cross-situational word learning, and second language acquisition will be discussed.

SpeakerHolden, John G.
Author 1Amon, Mary Jean
University of Cincinnati Psychology Department
Author 2Holden, John G.
University of Cincinnati Psychology Department
TitleCause, Context, and Replication in the Social Science
AbstractThe presence of two exotic patterns of variability in response time measures, 1/f noise and distribution scaling, have important and unanticipated implications for the replication crisis in the social and behavioral sciences. The core issue that both Frequentist and Bayesian inferential statistics share an identical and necessary foundational assumption. They require that statistical samples originate from populations representing static and effectively fixed parameters. The only logically admissible uncertainty in a classical inferential test is population parameter variability resulting from either deterministic and stable “signals”—potentially indicating distinct populations—or from unsystematic sources of additive “noise” indicating a relatively tame ergodic form of probabilistic uncertainty. Systems that entail scaling behavior routinely express dramatic parametric fluctuations and even nonlinearities. How then does one distinguish fluctuations from the logical change assumed by inferential techniques? Obviously, the discipline must retain inferential techniques. Their outcomes, however, must be suitably contextualized. Two empirical studies are presented to illustrate the issues at hand.

SpeakerJameson, Kimberly A.
Author 1Jameson, Kimberly A.
UC Irvine
TitleInterpersonal comparisons of color experience
AbstractColor is an inner, highly subjective, experience, initially triggered by properties of light from the external world. Actual color perceptions depend on (i) visual processing properties of observers that can vary greatly across individuals, and (ii) minor changes in viewing circumstances. This talk presents empirical results on visual processing behaviors of four individuals. All four of these individuals have excellent color perception (as shown by standardized color vision assessment procedures). Two of the individuals are considered standard “normal” trichromat observers, while the other two are “potential tetrachromat” observers – that is, observers with a genetic potential for an extra class of visual pigments used for color vision. By comparing such observers\' color perception performance, and some of their behavioral uses of color, we illustrate how wide the definition of “normal” human color vision actually is, and how some observers that are typically classified as “normal” might actually experience a world of color beyond what the average color vision normal human experiences.

Speakerjones, matt
Author 1jones, matt
university of Colorado
Author 2Zhang, Jun
University of Michigan
TitleDuality of similarity- and feature-based learning via kernel methods
AbstractThe kernel framework from machine learning offers a new perspective on psychological models of learning. In particular, recent work has shown that similarity-based generalization and feature-based association learning can be formally equivalent, provided the set of features bears the right relationship to the similarity function. Rather than treating this as an issue of model identifiability, we suggest viewing it as one of duality: The brain is doing similarity- and feature-based computation simultaneously. The kernel duality can be used to translate between these two modeling frameworks, using principles traditionally expressed in one to generate insights within the other. We illustrate this approach with the example of learned selective attention, showing how two very different theories of attention in learning -- grounded in similarity and in cue associability -- are complementary instantiations of the same general principle when cast within the kernel framework.

SpeakerKalish, Michael
Author 1Kalish, Michael
syracuse university
TitleTalking about category learning
AbstractCategory learning remains essentially an enigma. Different people do different things under different conditions. The primary method for determining who does what, when, is to describe individuals' response surfaces using a variety of models. These models are often identified as being indicative of particular strategies that participants might be using to learn the correct category labels for the stimuli. This is conceptually and methodologically problematic. Improved methodologies for estimating response surfaces may be helpful in understanding what people are doing. These methodologies may also illuminate essential limitations in the computational modeling of individuals' category learning processes.

SpeakerKetels, Shaw
Author 1Ketels, Shaw
University of Colorado, Boulder
Author 2Healy, Alice
University of Colorado, Boulder
Author 3Jones, Matt
University of Colorado, Boulder
Author 4Sassnett-Martichuski, Diane
University of Colorado, Boulder
Author 5Lalchandani, Lakshmi
University of Colorado, Boulder
Author 6Guhl, Mary
University of Colorado, Boulder
TitleExpertise reversal effects from variation in the use of classroom response systems.
AbstractWe manipulated the usage of classroom response systems, or “clickers,” in four statistics classes, taught by the same instructor at the University of Colorado Boulder. In two experiments each conducted over the course of a single semester, we evaluated two common pedagogical prescriptions for clicker usage: (a) Clicker questions should be interleaved throughout the class period, and (b) clicker questions should be presented with protected time for discussion. In both experiments, conditions were alternated within-subjects, and these patterns of alternation were counterbalanced between the two classes. Performance on midterm and final exam questions was used as the dependent measure for both experiments. For all students, superior test performance was expected for material presented on days with clicker questions interleaved throughout the class and days when students were given extra time to discuss. However, we found expertise reversal effects in both experiments: Interleaving questions and peer discussion both affected students differentially depending on their knowledge of, and/or exposure to, class material.

SpeakerKowler, Eileen
Author 1Kowler, Eileen
Department of Psychology, Rutgers University, Piscataway, NJ
Author 2Santos, Elio
Department of Biomedical Engineering, NJIT, Newark, NJ
TitleThe importance of prediction and expectations in eye movement control
AbstractThe ability to make accurate predictions about the future states of the world is critical for motor control and perception, including the control of movements of the eye (smooth or saccadic). Even though eye movement systems can process and react to sensory signals quickly, accurate predictive responses are nevertheless crucial for overcoming the harmful effects of processing delays. A role for expectation and prediction in eye movements is shown vividly by anticipatory smooth eye movements, which are involuntary smooth pursuit eye movements in the direction of expected future target motion. Expectations about the future path of a moving target can be derived from various signals, including visual cues that signal the likely upcoming motion path, as well as memory for previously seen target motions (Kowler et al., 2014, Journal of Vision; Santos, 2014). The strength of the anticipatory response depends on the perceptual qualities of the cue, the validity of the cue, and memory for recently seen motions. These recent results suggest that anticipatory smooth eye movements depend on two factors: the strength of the belief about the direction of future motion, and internal estimates of the costs of pursuit error. The study of anticipatory smooth eye movements can provide a window into how the brain formulates and uses predictions, an ability that is surely used broadly for perception and motor functions, and is not limited to the control of eye movements.

SpeakerLandy, David
Author 1Landy, David
Indiana University
Author 2Rogers, Brad
Indiana University
TitleThe probabilistic estimation of analogical relations
AbstractWhen faced with uncertain situations and when direct evidence is hard to come by, people often invoke information from structurally related categories or situations. People often draw relational correspondences (i.e., analogies) across instances drawn from different domains. People make these analogies when they struggle for purchase in uncertain situations, but nearly all prior psychological studies draw relational correspondences across knowledge structures that are treated as deterministic and error-free. We aim to understand relational structure mapping in contexts in which structured knowledge has quantified uncertainty. Further, extant approaches to relational correspondence have primarily considered the quality of a complex analogy between two domains to be a function of its structural consistency, invoking factors such as 1-1 correspondence, and the number and depth of structural correspondences. We present a theoretical framework that instead articulates the detection of relational correspondences normative estimation of the probability that the models generating uncertain data in the two domains share a particular type of relational correspondence: that is, we embed analogical inference in a model-testing framework. More probable analogies result when an observed relational structure match was particularly unlikely to occur in the absence of a particular correspondence holding across the true domain models, relative to a specified set of baseline models. Taking together the traditional approach and the probabilistic approach yields two criteria for a good analogy: degree of structural correspondence and the probability of correspondence of unknown generating models. We will demonstrate that some previously identified features of high-quality analogies align with properties that provide evidence for an analogy's 'truth', but that in general the two approaches make dissociable predictions. We will consider the structure-match and probabilistic models in the context of a novel experiment in which subjects make explicit statements about the probability of an analogy.

SpeakerLewandowsky, Stephan
Author 1Lewandowsky, Stephan
University of Bristol and University of Western Australia
Author 2Ballard, Timothy
University of Queensland and University of Bristol
Author 3Risbey, James
CSIRO Oceans and Atmosphere, Hobart, Tasmania
Author 4Brown, Gordon
University of Warwick
TitleHuman Wishful Thinking vs. Scientific Uncertainty as Knowledge: Constraints on Climate Policy Choices Provided by an Ordinal Analysis of Uncertainty
AbstractUncertainty forms an integral part of science, and uncertainty is intrinsic to many global risks that dynamically unfold over time, from “peak oil” to genetically modified foods to climate change. Uncertainty is often cited in connection with political arguments against corrective action. Using climate change as a case study, we report an ordinal analysis (i.e., statements of the form “greater than”) of uncertainty within the climate system. This analysis is not sensitive to people's cultural cognition or subjective risk perceptions and reveals that greater uncertainty (i.e., “greater than expected”) provides greater impetus for mitigative action. This normative result stands in contrast to people's tendency to view uncertainty as a stimulus for “wishful thinking”, and hence a reduced impetus for mitigative action. We explore the reasons underlying people's wishful thinking and suggest that, paradoxically, they may also reflect a normatively optimal adaptation to features of the environment. We examine the interplay between human cognition, physical reality, and policy options in a simulation experiment involving sea level rise.

SpeakerLittle, Daniel
Author 1Little, Daniel
The University of Melbourne
Author 2Wang, Tony
Brown University
Author 3Nosofsky, Robert
Indiana University
TitleRecency effects and response times in perceptual categorization: Comparing exemplar and rule-based accounts in a modified Garner task
AbstractA large number of converging operations suggesting that, unlike separable dimensions, integral dimensions are processed holistically. This difference has been convincingly demonstrated by Garner's (1974) classic study which showed that integral dimensions, but not separable dimensions, tend to interfere with each other if one of the dimensions must be ignored but tend to facilitate one another if the dimensions are varied in a correlated manner. One key aspect of Garner's results is that item and response repetitions resulted in faster response times. Here we report an experiment in which we increased the number of stimuli to reduce stimulus repetitions. Nonetheless we find clear recency effects in both response time and accuracy. We test three models of category choice response times including both exemplar-based and General Recognition Theory models, but also a modern version the distance from boundary theory in which utilizes an integrated array of linear ballistic accumulators. Model comparison results and the theoretical implications of the recency effects will be discussed.

SpeakerLove, Bradley
Author 1Love, Bradley
Author 2Kopec, Lukasz
TitleWhen Limits in Attention are Really Limits in Memory Retrieval
AbstractIn learning and decision tasks, people often bias attention in seemingly suboptimal ways, such as overweighting the most diagnostic cue at the expense of integrating across all cues. For example, in predicting which of two teams should win a game, people might focus on the star players to the exclusion of other predictive factors. The common interpretation is that capacity constraints in people's attentional system lead to suboptimal weighting. An alternative explanation is that the attentional system is compensating for the noise induced by a limited and stochastic memory retrieval process. This view follows from the observation that, when making a decision, people stochastically and selectively retrieve a small set of relevant memories that provide evidence for competing options. This limited retrieval injects harmful noise into the decision process. Previous work demonstrates that idealizing training information (i.e., overemphasizing clear cut cases and de-emphasizing ambiguous cases) improves performance by reducing the harmful effects of limited memory retrieval. In the current work, we find that people shift attention to "self-idealize" in a manner that reflects the individual's capacity limits in memory retrieval. Given memory retrieval limits, we find that attentional weighting is near optimal. Purported deficits of attention may in cases reflect limits in memory retrieval.

SpeakerMatzke, Dora
Author 1Matzke, Dora
Univerity of Amsterdam
Author 2Love, Jonathon
Author 3Heathcote, Andrew
TitleA Bayesian approach for estimating the probability of trigger failures in the stop-signal paradigm
AbstractResponse inhibition is frequently investigated using the stop-signal paradigm. In this paradigm, participants perform a two-choice response time task where the primary task is occasionally interrupted by a stop signal that instructs participants to withhold their response. Stop-signal performance can be formalized as a race between a go process that is initiated by the primary task stimulus and a stop process that is triggered by the stop signal. If the go process wins, the primary response is executed; if the stop process wins, the primary response is inhibited (Logan & Cowan, 1984). Successful response inhibition requires relatively fast stop responses as well as a high probability of triggering the stop process. Existing methods allow for the estimation of the latency of the stop response, but are unable to identify deficiencies in triggering the stop process. We introduce a Bayesian mixture model that addresses this limitation and enables researchers to simultaneously estimate the probability of trigger failures and the entire distribution of stopping latencies. We demonstrate that trigger failures play an important role in the stop-signal performance of healthy participants, and that ignoring them distorts estimates of stopping latencies. Moreover, we introduce BEESTS-WTF, a user-friendly software implementation of our trigger-failure framework. BEESTS-WTF comes with a graphical user interface and provides users with summary statistics of the posterior distribution of the parameters as well various diagnostics tools to assess the quality of the parameter estimates. The software is freely-available and runs on OS X and Windows operating systems.

SpeakerMueller, Shane
Author 1Mueller, Shane
Michigan Technological University
Author 2Thanasuan, Kejkaew
Michigan Technological University
TitleModeling the sources of fluent expert memory access and search in competitive crossword players
AbstractCrossword play requires memory along two routes (semantic and orthographic) that provide complementary cues and constraints. The process can be understood as a memory search problem in which candidates are generated via memory retrieval, and then checked against the constraints to determine whether the candidate is satisfactory. It remains an open question whether this memory search can happen simultanously along both routes, or must happen for orthographic and semantic routes separately. We report the results of experimentation and a computational model that show the best explanation, for both novices and experts, is that memory search occurs with one type of cue at a time. This suggests that compound cues are typically not used to search memory in crossword play. Furthermore, these results, together with a computational model of crossword play, indicate that expert players may be especially adept at memory access via semantic (clue-answer) associations. This suggests an association-based account of knowledge expertise wherein recognitional decisions are constrained mainly be fluent memory retrieval rather than a more traditional decision process involving the ability to compare and weigh between options.

SpeakerMyung, Jay
Author 1Myung, Jay
Ohio State University
TitleA hierarchical Bayesian approach for optimized adaptive experiments
AbstractExperimentation is at the core of research in the behavioral and neural sciences, yet observations can be expensive and time-consuming to acquire (e.g., MRI scans, responses from infant participants). A major interest of researchers is designing experiments that lead to maximal accumulation of information about the phenomenon under study with the fewest possible number of observations. In addressing this challenge, our lab has developed an adaptive design optimization (ADO) method. In this talk, I present its hierarchical Bayes extension, dubbed hierarchical ADO (HADO), that provides a judicious way to exploit two complementary schemes of inference (with group and individual data) to achieve even greater accuracy and efficiency in information gain. Also discussed are results from a validation study to evaluate the benefits and validity of HADO in both human and simulation experiments in the adaptive estimation of the contrast sensitivity function in visual psychophysics.

SpeakerOsth, Adam
Author 1Osth, Adam
University of Newcastle
Author 2Dennis, Simon
University of Newcastle
TitleSources of Interference in Item and Associative Recognition Memory
AbstractA powerful theoretical framework for exploring recognition memory is the global matching framework, in which a cue's memory strength reflects the similarity of the retrieval cues being matched against the contents of memory simultaneously. Contributions at retrieval can be categorized as matches and mismatches to the item and context cues, including the self match (match on item and context), item noise (match on context, mismatch on item), context noise (match on item, mismatch on context), and background noise (mismatch on item and context). We present a model that directly parameterizes the matches and mismatches to the item and context cues, which enables estimation of the magnitude of each interference contribution (item noise, context noise, and background noise). The model was fit within a hierarchical Bayesian framework to ten recognition memory datasets that employ manipulations of strength, list length, list strength, word frequency, study-test delay, and stimulus class in item and associative recognition. Estimates of the model parameters revealed at most a small contribution of item noise that varies by stimulus class, with virtually no item noise for single words and scenes. Despite the unpopularity of background noise in recognition memory models, background noise estimates dominated at retrieval across nearly all stimulus classes with the exception of high frequency words, which exhibited equivalent levels of context noise and background noise. These parameter estimates suggest that the majority of interference in recognition memory stems from experiences acquired prior to the learning episode.

SpeakerPauli, Wolfgang (spin +1/2)
Author 1Pauli, Wolfgang (spin +1/2)
Caltech, Pasadena, USA
Author 2O\'Reilly, Randall C.
University of Colorado Boulder, CO, USA
Author 3Yarkoni, Tal
UT Austin, TX, USA
Author 4Wager, Tor
University of Colorado Boulder, CO, USA
TitleRegional specialization within the human striatum: novel insights from an unbiased data-driven approach.
AbstractDecades of animal work and recent human neuroimaging analyses have identified distinct, but overlapping, striatal zones interconnected with different cortical and thalamic circuits. These zones are crucial for the organization of functional systems. Despite continuous efforts to subdivide the human striatum based on anatomy and resting-state connectivity, however, characterizing the different psychological processes related to each zone remains a work in progress. Here, we followed a data-driven approach, and analyzed large-scale co-activation data from nearly 6,000 human imaging studies. Following classical anatomical taxonomies, we (a) identify five distinct striatal zones with discrete patterns of co-activation with distal brain regions across many psychological conditions, and (b) characterize the different psychological processes associated with these zones. We find that the reported pattern of cortical activation in a study can be used to predict selective activation of striatal zones, and that activation in each functional zone is associated with distinct psychological processes. Some of these associations are well-established, such as the role of the ventral striatum in reward processing. Others are less established, such as a (1) selective involvement of the ventral striatum and adjacent anterior caudate in evaluating the value of rewards and actions, respectively, and (2) a specialization of the posterior caudate nucleus for executive functioning, often considered to be the exclusive domain of the prefrontal cortex. Our findings highlight the strong regional specialization within the human striatum for different psychological tasks and provide a unique view into underlying task demands.

SpeakerPecher, Diane
Author 1Pecher, Diane
Erasmus University Rotterdam
Author 2Roest, Sander
Erasmus University Rotterdam
Author 3Stötefalk, Nic
Erasmus University Rotterdam
Author 4Fiere, Alma
Erasmus University Rotterdam
Author 5Zeelenberg, René
Erasmus University Rotterdam
TitleAction potentiation by object pictures
AbstractResponses to pictures of graspable objects are influenced by the similarity between the response action itself and the actions that could be performed with the object. Thus, pictures of objects potentiate object-related actions. 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 with the hand that is closest to the handle. When the response involves the same hand or the same grip, responding is facilitated compared to a different hand or different grip because the response action is already activated by the object picture. 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, for example spatial location. When the dimensions are aligned, responses are faster than when they are not aligned, as in the Simon effect. In several paradigms, we found that spatial attention and the presence of response competition could explain performance. This suggests that stimulus-response competition is a more likely explanation than sensory-motor simulations.

SpeakerPothos, Emmanuel
Author 1Pothos, Emmanuel
City University London
Author 2Yearsley, James
City University London
Author 3White, Lee
Swansea University
TitleConstructive influences in decision making: a quantum perspective
AbstractIt is well known that, in some cases, a judgment or choice can affect the underlying mental representations; that is, judgments or choices can have a constructive influence. Constructive influences can be fairly easily incorporated in standard cognitive models, for example, through assumptions that a judgment alters memory or attention for the related information. Interestingly, if one attempts to model (some aspects of) cognition using quantum theory, then certain judgments are required to be constructive. This is because so-called superposition states in quantum theory are such that a value for a corresponding observable does not exist prior to a measurement; rather, the measurement creates the observed value. In cognition, if certain opinion states are like superposition ones, then a judgment would likewise create the representations, consistent with the outcome of the judgment. Constructive influences in quantum models are thus a deep, structural feature of such models. Moreover, constructive influences have to be of a very specific kind. We present a simple experimental situation, whereby a second stimulus is always rated, but a previous, first stimulus is sometimes rated, sometimes not. A corresponding quantum model for how the first rating might impact on the second one is developed and its predictions confirmed across several experiments.

SpeakerRatcliff, Roger
Author 1Ratcliff, Roger
The Ohio State University
Author 2McKoon, Gail
The Ohio State University
TitleNumeracy, Aging, and Individual Differences
AbstractThirty one elderly adults were tested on 5 numeracy tasks: a symbolic task (is this 2-digit number greater or less than 50), a non-symbolic task (is the number of asterisks in this array greater or less than 50), a go/no-go version of the non-symbolic task, a task used to control for brightness and area, and a number memory task. The diffusion model was fit to the data and model parameters were compared between the elderly adults and college-age adults. Individual differences in model parameters across the tasks (and IQ) were also compared.

SpeakerShiffrin, Richard
Author 1Shiffrin, Richard
Indiana University
Author 2Cao, Rui
Indiana University
TitleFinding targets is faster than finding foils
AbstractAfter study of a list of words, we test with target search: pick out the one list- word presented with three non-list-words, or test with foil search: pick out the one non-list-word presented with three list-words. Target and foil search are equally accurate but target search is much faster. The same is true when two choice words are presented. We suggest that target strengths are more strongly positively skewed than are foils, since only targets are studied and study sometimes produces large trace strengths. We modeled the effects with a race between four ballistic evidence accumulators. We discuss other (less plausible) models that can also predict these findings. The results and models should therefore be viewed as a guide for future research.

SpeakerSikström, Sverker
Author 1Sikström, Sverker
Department of Psychology, Lund University, Sweden
Author 2Kempe, Tomas
Department of Psychology, Lund University, Sweden
Author 3Söderlund, Göran
Sogn og Fjordane University College
TitleFrequency Dependency in Stochastic Resonance
AbstractThe detection of a signal can be enhanced by the addition of noise in systems that are dependent on a threshold. This stochastic resonance phenomenon has previously been described to be fully dependent amplitude of the noise and the signal. We show theoretically that another factor, namely the signal's frequency, to a large extent modulates the likelihood of detecting the signal, and where detection is facilitated at low frequencies where more samples allows a better estimate of the noise. The theory was tested in an auditory signal detection task where low frequency (500 Hz) signals benefited more from noise than high frequency signals (4000Hz), providing support for the proposed theory. Furthermore a new and promising method for optimising the signal-to-noise ratio during the test session where evaluated. This method is unique compared to earlier studies in the sense that we kept the signal to noise ratio (SNR) between noise and stimuli constant during the conditions so that the noise volume parallel the stimuli volume during the sessions.

SpeakerSperling, George
Author 1Sperling, George
University of California, Irvine
Author 2Chubb, Charles
University of California, Irvine
Author 3Wright, Charles E. (Ted)
University of California, Irvine
Author 4Sun, Peng
New York University, NY NY
Author 5Inverso, Matthew
University of California, Irvine
Author 6Ton, Pauline
University of California, Irvine
Author 7Blair, Garrett
University of California, Irvine
Author 8Winter, Nicole
University of California, Irvine
TitleParadoxical anomalies in centroid SSRs
AbstractA centroid judgment (clicking the center of gravity of a cloud of items) is a Statistical Summary Representation made following a briefly exposed visual display. Centroid judgments are useful in studying early processes of selective feature-attention. Subjects find the centroid of only a designated subset of items while ignoring the remainder, e.g., finding the centroid of items lighter than the background while ignoring items darker than the background. At ASIC-2014, I reported in great detail on subjects' abilities to selectively attend to a requested color while ignoring other colors, and the perceptual attention filters for color derived therefrom. Time permitting, I will review three recent paradoxical anomalies.
   (1) Whereas subjects can accurately judge centroids of dark versus light dots, large versus small squares, squares versus triangles, and many other features, they cannot judge the centroid (nor the numerosity) of vertical bars among horizontal bars or vice versa.
   (2) Subjects can find the centroid of the more numerous dots, e.g., 12 dots of one color among 2 dots of each of 7 other colors, and vice versa, without any prior knowledge of the dot colors which vary on each trial.
   (3) In some cases, subjects are more efficient and more accurate in finding the centroid of items defined by a conjunction of features than in finding centroids of items defined by a unique feature. This is very different from results in visual search.
   Together, these results suggest that the processes involved in centroid judgments--and other SSRs--are different from previously-studied attentional processes.

SpeakerStarns, Jeffrey
Author 1Starns, Jeffrey
UMass Amherst
Author 2Staub, Adrian
UMass Amherst
Author 3Chen, Tina
UMass Amherst
TitleImplications of response time and eye movement data for models of forced choice recognition
AbstractIn a forced-choice recognition task, participants are asked to select which of two words appeared on an earlier list. Researchers have claimed that forced-choice tests can be used to discriminate signal detection models that make different assumptions about the nature of memory retrieval, and forced-choice recognition has also played a role in constraining process models of memory. Throughout the forced choice literature, models assume that participants compare the memory strength of the two items and select the one with stronger memory evidence. We evaluated response time (RT) and eye movement data in two forced-choice recognition experiments. Results suggest that participants base decisions on the absolute memory evidence for each alternative, and they only resort to evaluating the relative evidence when they are uncertain of the correct response. I will discuss how these results constrain potential sequential sampling models for forced-choice accuracy and RT data.

SpeakerStephens, Rachel
Author 1Stephens, Rachel
Syracuse University
Author 2Dunn, John
University of Adelaide
Author 3Hayes, Brett
University of New South Wales
Author 4Kalish, Michael
Syracuse University
TitleThe influence of logic training and believability on inductive and deductive judgments
AbstractDual-process accounts posit that separate heuristic and analytic processes contribute to reasoning. Under this view, inductive judgments are more heavily influenced by the quick heuristic processes that use background knowledge or associative information. In contrast, deductive judgments are more strongly influenced by the slower analytic processes that are more deliberate and rule-based. However, our recent meta-analysis of existing research showed there is limited evidence that this complex account is required. Rather, a simpler single-process theory can account for both inductive and deductive judgments. Guided by our meta-analysis, we conducted two new experiments in search of evidence of the dual process account. Crucially, two factors were manipulated that might be expected to differentially affect heuristic and analytic processes. Participants judged the strength of written valid and invalid arguments, with separate groups using either inductive or deductive criteria. We factorially manipulated whether the arguments were believable according to background knowledge (which should have a greater influence on heuristic processes), and whether participants had received training on how to correctly assess logical validity for the difficult argument forms (which should have a greater influence on analytic processes). We found that though both factors indeed influenced people's inductive or deductive judgments, there was still no evidence that the single-process account should be rejected in favor of the dual process account.

SpeakerSteyvers, Mark
Author 1Steyvers, Mark
UC Irvine
Author 2Merkle, Ed
University of Missouri
Author 3Mellers, Barbara
University of Pensylvania
Author 4Tetlock, Philip
University of Pensylvania
TitleLearning about movie preferences and forecasting abilities in the presence of missing data
AbstractMissing data is a common problem in modeling and statistical inference. To analyze and learn from incomplete data, assumptions need to be made about the data generating process that generates complete data and the missing data process that explains which elements of the complete data will not be observed. We will present two case studies of where it is important to include choice processes as part of the missing data process. The first case study is in the context of recommender systems (e.g. Netflix, Goodreads) where the goal is to learn about user preferences from a set of items (e.g. movies, books) that are rated by the user. The items to be rated are chosen by the user and are not randomly picked by an experimenter. We will present a topic model that includes a probabilistic process for the choice of items to rate as well as a process for selecting a rating for those items. These topic models can learn about user preferences from just knowing which items were rated even in the absence of any explicit ratings. The second case study is based on a large-scale forecasting tournament where users choose the forecasting problems they work on and for each problem, they provide a probabilistic forecast. We show how an Item Response Theory (IRT) can be generalized to not only handle continuous probabilistic forecasts but also the choice of items for which users forecast. We show how the selected forecasting problems from each forecaster provides a new source of data for estimating forecaster ability. Generally, the two case studies suggest that missing data due to choice processes present computational challenges (how do we model the choice processes) but also opportunities to gain additional information

SpeakerTeodorescu, Kinneret
Author 1Plonsky, Ori
Technion - Israel Institute of Technology
Author 2Teodorescu, Kinneret
Indiana University
Author 3Erev, Ido
Technion - Israel Institute of Technology
TitleReliance on small samples, the wavy recency effect and similarity-based learning
AbstractMany behavioral phenomena, including underweighting of rare events and probability matching, can be the product of a tendency to rely on small samples of experiences. Why would small samples be used, and which experiences are likely to be included in these samples? Popular learning models assume reliance on the most recent experiences due to cognitive limitations and/or adaptation to gradually changing environments. We explore a very different and more cognitively demanding process explaining the tendency to rely on small samples: exploitation of environmental regularities. In computational analyses we show that across wide classes of environments, focusing only on experiences that followed the same sequence of outcomes preceding the current task is more effective than focusing on the most recent experiences. We then examine the psychological significance of these sequence-based models. Most learning models predict that the impact of each outcome will be maximal immediately after its occurrence, and will diminish monotonically with time (positive recency). In contrast, sequence-based rules predict a non-monotonic development over time with three distinct stages: The initial effect is negative, then it becomes positive, and finally, in the long term, the effect diminishes. Analysis of published data supports this non-trivial wavy recency pattern and shows robust sequential dependencies ignored by previous research. For example, the tendency to underweight a rare event is found to be strongest three trials after its occurrence. Thus, despite their cognitive cost, sequence-based models have appealing descriptive value. Implications to similarity-based learning and learning models in general will be discussed.

SpeakerTeodorescu, Andrei
Author 1Teodorescu, Andrei
Psychological and Brain Science, Indiana University
TitleFalsifying unfalsifiable models - grounding model inputs in stimulus values rather than free parameters
AbstractIn 2013, two papers were published in Psychological Review pointing out the crucial problem of model mimicry within the sequential sampling model class and its origin in arbitrary technical model assumptions. However, while the work by Jones & Dzhafarov concludes that the entire class is unfalsifiable, the work by Teodorescu & Usher endeavors to provide a framework of theoretically driven experimental design which generates non-overlapping, and thus falsifiable, predictions from different models. How can such opposed conclusions co-exist? In this talk I will try to bridge the two works by discussing the similarities and differences and illustrate the value of emerging insights in a follow up study to Teodorescu & Usher (2013). The hallmark of experimental psychology has been the use of clever experimental designs to flush out differences between theories and compare their predictions. The hallmark of cognitive modeling has been the use of clever mathematical and computational models that can account for the experimental results. In the void between these two enterprises lay assumptions that are necessary for linking models with experiments. The most fundamental of these are the “selective influence” assumptions which confine the effects of a particular manipulation to a unique set of model parameters. Common practice allows parameters to vary freely with their selectively influencing experimental manipulations. I will try to argue and demonstrate that this freedom is excessive and that constraining momentary model input values to momentary stimulus values can allow us to avoid model mimicry and improve model selection.

Speakervan Ravenzwaaij, Don
Author 1van Ravenzwaaij, Don
University of Newcastle
Author 2Brown, Scott
University of Newcastle
Author 3Marley, Anthony
University of Victoria
Author 4Heathcote, Andrew
University of Newcastle
TitleThe Advantage Linear Ballistic Accumulator: A New Model for Multi-Alternative Forced Choice Tasks
AbstractOver the last few decades, cognitive psychology has seen an advent of sequential accumulator models that aim to fit response time data from forced choice tasks. When the number of response options is higher than two, these models tend to posit one accumulator per response option: evidence accumulation is conceptualized as absolute evidence for one response option. Here, we propose a new model for sequential evidence accumulation in which evidence is collected relative to other response options: the advantage linear ballistic accumulator. In the first part of this paper, we present three kinds of model architectures that differ in terms of the conditions that have to be met for a response to be chosen. We demonstrate in model simulations that all of these architectures naturally produce Hick's Law (Hick, 1952). In the second part, we present fits of one model architecture (the Win-All model) to an empirical Hick's Law dataset. In the third part of the paper, we discuss a recent claim by (Teodorescu & Usher, 2013), that in order to account for some empirical multi-alternative forced choice data, sequential accumulator models need mutual inhibition. We present fits of the Win-All model that does not include mutual inhibition to data by Teodorescu and Usher (2013).

SpeakerVul, Ed
Author 1Vul, Ed
UCSD Psychology
Author 2Lew, Timothy F
UCSD Psychology
TitleHow visual working memory exploits environmental structure.
AbstractHow do people use the structure of items when storing them in visual memory? Experiment 1 asked what format visual working memory uses to encode objects and their structure. Subjects saw objects arranged in different spatial clustering structures and recalled their positions. Objects in the same cluster were misreported in similar directions, indicating that memory errors were shared within clusters. Additionally, the shared errors for clusters decreased when clusters were closer. These results are captured by a model that encodes object positions relative to an inferred grouping structure and recalls relative positions with Weber noise. Experiment 2 adopted an iterated learning paradigm to amplify biases due to people's prior expectations about spatial structure. Each subject saw 15 items and reported their positions; critically, the positions one subject reported served as the stimulus for the next subject. People converged to reporting items in few groups that are either tight clusters or lines, and multiple lines in a display with similar orientations and lengths. This effectively recovers visual memory's use of Gestalt principles to encode objects. Together, these results show how people use environmental structure to remember displays: what structures they expect and exploit, and what format encodes objects and their structure.

SpeakerWagenmakers, Eric-Jan
Author 1Wagenmakers, Eric-Jan
University of Amsterdam
TitleStatistical Inference for Irrelevant Data
AbstractWhen two models make identical predictions for a particular set of observations, this set is called irrelevant, as it cannot be used to discriminate the models. Based on earlier work by Harold Jeffreys, I demonstrate that the first binomial observation is irrelevant for discriminating H0 (theta = 1/2) from H1 (theta ~ symmetric around 1/2). Then I demonstrate that the nth binomial observation is likewise irrelevant, provided that the n-1 previous observations are split evenly between successes and failures. Finally, I generalize the concept of irrelevance by introducing Maximally Uninformative Data (MUD) sequences; given a particular prior distribution, there exists a matching MUD sequence, that is, an infinitely long sequence of observations for which the predictive adequacy of H0 equals that of H1 throughout. Interestingly, all MUD sequences will yield p < alpha for any alpha, and all MUD sequences will produce confidence intervals that do not overlap with the parameter under test. This behavior can be explained by interpreting the p-value in Bayesian terms.

SpeakerZeelenberg, Rene
Author 1Zeelenberg, Rene
Erasmus University Rotterdam
Author 2Pecher, Diane
Erasmus University Rotterdam
TitleThe Role of the Motor System in Short-Term and Long-Term Memory for Objects and Words
AbstractIt has been suggested that action is central to cognition. Recent studies suggest that actions may be automatically activated by objects and words (e.g., Tucker & Ellis, 2001), but little is known about the role of the motor system in short-term and long-term memory. Shebani and Pulvermüller (2013) recently reported evidence supporting a role for motor simulations in immediate serial order recall for words. In their study, movements with the hands impaired working memory for arm-related action words (e.g., grab, stir) and movements with the feet impaired working memory for leg-related action words (e.g., kick, skate). We report several experiments that investigated whether similar effects are present in other short-term and long-term memory tasks.

SpeakerZhang, Shunan
Author 1Zhang, Shunan
Author 2Song, Amanda
Author 3Yu, Angela
TitleA Bayesian hierarchical model of crowding: a case study of global-local processing in visual perception
AbstractWe explore the interaction between global-local information processing in visual perception, using a visual phenomenon known as crowding, whereby the perception of a target stimulus is impaired by the presence of nearby flankers. The majority of established models explain the crowding effect in terms of local interactions. However, recent experimental results indicate that a classical crowding effect, the deterioration in the discrimination of a vernier stimulus embedded in a square, is alleviated by additional squares (\"uncrowding\"). Here, we propose that crowding and uncrowding arise from abstract neural inferences about hierarchically organized groups, and formalize this hypothesis using a hierarchical Bayesian model. We show that the model reproduces both crowding and uncrowding; more generally, the model provides a normative prescription for how visual information might show bottom-up, top-down, and laterally, to allow the visual system to simultaneously and interactively process global and local features in the visual scene.

SpeakerZhang, Byoung-Tak
Author 1Zhang, Byoung-Tak
Seoul National University
TitleReverse Engineering the Embodied Mind by Human Robotics
AbstractBehaviorism has focused on measurable stimulus-response relationships of human behavior while ignoring cognition. In contrast, cognitivism has focused on the internal information processing mechanisms of the mind while ignoring the body and action. Recent studies in cognitive science emphasize the embodied mind and its interaction with the environment within the perception-action cycle. However, many researchers believe that, despite its significance, the progress of the embodied and situated mind research would be quite slow due to the technical difficulties of sensing and modeling the experimental data in real worlds. In this talk we argue that the emerging wearable technology, such as smart glasses and wearable EEG devices, and machine learning come to the rescue. Based on this idea, we present a new research paradigm for reverse engineering the embodied mind (i.e. understanding human thoughts and acts) in ecologically-valid environments using wearable devices and robotics technology. The proposed “human robotics” approach to cognitive science views the wearable devices as robots that continually sense and track the everyday activity of the wearers (a “wearable robotics” problem from the robotics point of view). Using machine learning technology combined with mobile and cloud computing, the wearable robots attempt to reproduce or “clone” the human mind and behavior in real-world in real-time over an extended period. We take the mindcloning problem as an example to illustrate the human robotics paradigm and discuss its experimental setups, applications, prospects, and the challenges in cognitive modeling.