Skip navigation and jump directly to page content
Indiana University Bloomington
Choose site to be searched
Type search terms

Cognitive Lunch

The Cognitive Lunch talks will be on Wednesdays from 12:10 pm - 1:25 pm in the Psychology conference room (PY 128) located behind the main office.

Fall 2014 Cognitive Lunch

  • Aug 27, 2014 - Tom Busey
  • Sep 10, 2014 - John Searle
  • Sep 17, 2014 - Ehtibar Dzhafarov (Purdue University)
  • Sep 24, 2014 - Yanjiang Wang
  • Oct 1, 2014 - Jon Willits
  • Oct 8, 2014 - Partric Simen (Oberlin College)
  • Oct 15, 2014 - Jonathan Tullis
  • Oct 22, 2014 - Sven Bambach
  • Oct 29, 2014 - Maxim Bushmakin
  • Nov 5, 2014 - Cindy Hmelo-Silver
  • Nov 12, 2014 - Fermin Moscoso del Prado Martin (UCSB)
  • Dec 3, 2014 - Jared Lorince
  • Dec 10, 2014 - Cotie Long


Aug 27, 2014: Tom Busey
Title: Tracking The Growth of Evidence in Visual Comparison Tasks
Abstract: Many applied tasks involve a comparison between visual images. For example, the Boston Bombing case was solved primarily through the manual comparison of images of the bombers with faces from other databases. Of course, no two picture or impression are identical, and thus the examiner must make judgments about the likelihood of the two images deriving from the same source. In the present work we collected data from fingerprint examiners when they performed a visual comparison task not unlike a latent print examination. In the first stage we asked examiners to select regions they thought might be diagnostic. In the second stage we showed these regions one at a time and asked for a judgment of whether the two came from the same source or different sources. We fit models based on signal detection theory to summarize the results. We found that, contrary to a model of holistic or configural processing, evidence seems to grow linearly with the number of regions. We also can track the growth of evidence in favor of exclusion or identification, and I'll present data on this point. The design addresses how weak and strong evidence is interpreted and has some surprising results about how weak evidence is evaluated.

Sep 10, 2014: John Searle
Title: Changing Paradigms in Cognitive Science

Sep 17, 2014: Ehtibar Dzhafarov (Purdue University)
Title: Probabilistic Contextuality: A general theory
Abstract: Probabilistic Contextuality is a notion primarily used in foundations of quantum mechanics, but it is applicable to all areas of research where systems with random outputs are recorded in response to deterministic inputs. Psychology, with its response variables almost always random, is one of these areas. The meaning of Probabilistic Contextuality relates to two foundational issues in the Kolmogorovian Probability Theory: the issue of identity of random variables (as distinct from their distributions), and the issue of “sewing together” (probabilistically coupling) random variables conditioned on different, mutually incompatible conditions. In a nutshell, contextuality is being brought in as follows. One hypothesizes that input x affects only output A and input y affect only output B. This hypothesis may be upheld or rejected by means of one of numerous tests (in psychology, they are called tests for selective influences, and they include, as a subset, the tests used in quantum mechanics). If the hypothesis of selectiveness is wrong, there are two possibilities: either the violations of selectiveness can be explained by “direct” cross-influences (e.g., A is influenced by both x and y “directly”), or this is not the case. This latter case is the one when we speak of Probabilistic Contextuality. I will discuss new developments that allow one to define and measure both direct cross-influences and contextuality.

Sep 24, 2014: Yanjiang Wang
Title: Inferring Structural Brain Connectivity Using Resting-state fMRI by Network Deconvolution
Abstract: The human brain is highly active all the time, with signals propagating between sets of brain regions along structural pathways (e.g. the cortical white matter). fMRI activation studies identify regional task-related changes in the hemodynamic response that are superimposed on significant levels of baseline or “intrinsic” brain activity. Diffusion tensor imaging (DTI) and tractgraphy are widely used to map structural connections. Numerous studies have examined correlated activity among brain regions measured in a task-free “resting-state”. These show that the strongest positive correlations appear among regions connected by the strongest and densest structural pathways. Systematic comparisons of structural and resting-state functional connectivity have had significant success in using the observed DTI structural strengths to predict the fMRI correlations, including correlations among regions that are only indirectly structurally coupled. For example, one approach uses a communication model where fMRI correlations are predicted based on the topology of the shortest (and presumably most efficient) structural paths (O. Sporns, et al. PNAS, 2014). The present research shows how one might go in the other direction. We employ a deconvolution model that accounts for direct and indirect effects and can predict structural connectivity from fMRI correlations among the brain regions. The deconvolution model assumes (for computational practicality) that fMRI correlations are produced by the sum of one and two step structural paths, and finds the best structure that produces the observed fMRI correlations. This deconvolution uses a biologically inspired high dimensional matrix optimization algorithm (termed cell differentiation algorithm, or CDA) and can produce predictions for structural connections among 500 regions. The resultant predictions are filtered further by keeping connections whose shortest paths are strong. We apply the model and the algorithm to one high-resolution dataset (998 cortical regions, 5 subjects) and one low-resolution dataset (90 cortical regions, 8 subjects). The results show that most of the inferred intrahemispheric structural connections are in agreement with the empirical results by DTI for both cases, while some missed interhemispheric connections that are hard to measure with DTI can be uncovered as well. Simulated FC by neural mass modeling supports the finding.

Oct 1, 2014: Jon Willits
Title: What Can Children Learn from 6 Million Words?
Abstract: What could children learn about words’ meanings from their distributional statistics? What do the statistics tell us about the possible organization of semantic memory? We addressed these questions by constructing a semantic model using the distributional statistics for 6,000,000 words of child-directed speech (from the CHILDES corpus), spoken to children under five years of age. First, we assessed the usefulness of the model’s representations for inferring nouns’ category memberships. The model’s average performance was 83.5%% correct (ranging from 70-91%, depending on the category). Second, we investigated the hierarchical nature of the model’s representations, by comparing words’ between- and within- category similarity to other words at various taxonomic levels. We found that the semantic space that emerges from the distributional statistics is strongly hierarchically organized. These analyses demonstrate that small samples of distributional statistics are useful for inferring semantic relationships such as words’ category memberships, and that these statistics suggest that taxonomic and hierarchically structured representations ought to emerge naturally, given the structure of the input.

Oct 8, 2014: Partric Simen (Oberlin College)
Title: Interval Timing and Decision Making with an Opponent Poisson Diffusion Model
Abstract: Diffusion models explain patterns of choice probabilities and response times in perceptual decision tasks. In these models, a random quantity representing "evidence" builds up over time until a boundary is reached, yielding a decision. Recently, my colleagues and I have applied diffusion models to a different task: interval timing. Despite major differences between these task types, response time distributions in both typically exhibit constant coefficients of variation (CVs) across conditions. To understand how diffusion could explain CV invariance, we analyzed diffusion processes that emerge from accumulating excitatory and inhibitory Poisson spikes. Assuming spike rates are proportional to evidence strength (in decision making) or clock speed (in timing) leads to CV invariance and conformance to Weber's law under certain changes of task conditions, but not others. We tested these predictions in three sensory modalities. Most held up, suggesting that decision making and timing may derive from a common process of diffusion.

Oct 15, 2014: Jonathan Tullis
Title: That reminds me: The influence of unplanned retrievals on memory and interpretation
Abstract: We often navigate novel situations by utilizing relevant past episodes. Remindings – stimulus-guided retrievals of prior episodes – allow us to link prior knowledge to current problems by prompting retrieval of relevant knowledge from past events. Here, I explore the consequences of remindings on memory of related material and interpretation of ambiguous information. I show that remindings benefit the recollection of the retrieved episodes and bias the interpretation of indefinite stimuli. Implications for theories of remindings will be discussed.

Oct 22, 2014: Sven Bambach
Title: Tracking and Disambiguating Hands in Egocentric (First-Person) Video
Abstract: Egocentric cameras are becoming more popular, introducing increasing volumes of video in which the biases and framing of traditional photography are replaced with those of natural viewing tendencies. This paradigm enables new applications, including novel studies of social interaction and human development. Recent work has focused on identifying the camera wearer’s hands as a first step towards more complex analysis. Inspired by the idea of better understanding how different hands contribute towards visual attention in toddlers during child-parent interaction, we study how to disambiguate and track not only the egocentric observer’s hands but also those of social partners. We present a probabilistic computer vision framework for modeling paired interactions that incorporates the spatial, temporal, and appearance constraints inherent in egocentric video.

Oct 29, 2014: Maxim Bushmakin
Title: Cognitive Lunch

Nov 5, 2014: Cindy Hmelo-Silver
Title: Cognitive Lunch

Nov 12, 2014: Fermin Moscoso del Prado Martin (UCSB)
Title: Cognitive Lunch

Dec 3, 2014: Jared Lorince
Title: Cognitive Lunch

Dec 10, 2014: Cotie Long
Title: Cognitive Lunch

Previous Cognitive Lunch