Indiana University Bloomington










  • 01/14    Organizational - Abstract
  • 01/21    Rich Shiffrin - Abstract
  • 02/04    Thomas W. James, Department of Psychological and Brain Sciences, Indiana University - Abstract
  • 02/18    Jun Zhang, Department of Psychology, University of Michigan, Ann Arbor - Abstract
  • 02/25    Joe Houpt - Abstract
  • 03/04    Dan Fogerty - Abstract
  • 03/11    Skyler Place - Abstract
  • 03/25    Cameron Buckner - Abstract
  • 04/01    John K. Kruschke - Abstract
  • 04/08    Stephen E. Denton and John K. Kruschke - Abstract
  • 04/15    Leslie Blaha - Abstract
  • 04/22    Paul Williams - Abstract
  • 04/29    Douglas Hofstadter - Abstract
  • 05/06    Benjamin Scheibehenne - Abstract

Abstract

1/14:    Organizational


1/21:    Rich Shiffrin
Model Selection for Dummies

Model selection in the narrow sense is best viewed as statistical inference, although there are many larger concerns as well. Although a non-expert, I have been involved with many of the modern developments in this area, and even published (with Andrew Cohen and Adam Sanborn) a recent PB&R article on the subject. Access to this field for non-experts is a daunting task, due to technical gimcrackery, uninterpretable terminology, non-stop argumentation among experts about the 'right approach', and mixtures of philosophical, mathematical, and empirical justifications for the alternative approaches. Yet statistical inference in the modern age absolutely requires the use of these modern approaches. Thus scientists are increasingly using model selection techniques, mostly in the form of such simple approximations as AIC and BIC, often without understanding the inference issues that are involved. Even experts can lose their way and lose sight of the basic underlying conceptual issues, and non-experts often cannot find their way at all. In this cognitive lunch talk I will discuss some of the larger model selection issues, but spend most of the time on the narrow issue of statistical inference. The exposition will be almost entirely non-technical (certainly by the standards of this field), but will nonetheless focus on the two or three leading modern approaches to model selection. I hope to illuminate the core conceptual issues. (There will be ample room for experts to express their horror at any mis-characterizations I introduce, but with luck, even experts may find a few nuggets of wisdom in the discussion).

2/4:    Thomas W. James, Department of Psychological and Brain Sciences, Indiana University
Neural Substrates of Visual-Haptic Object Recognition

In humans and many other primates, vision plays the major role in object recognition. But, objects can also be recognized by touch. In fact, visual and tactile (or haptic) object recognition systems have in common the ability to recognize and/or represent objects based on their volumetric shape. The object recognition literature, both behavioral and neural, is dominated by studies of vision, so why is haptic object recognition research important? Besides determining the object recognition capabilities of the haptic system itself, studying haptic object recognition also helps constrain theories of object processing in general. Real-world object recognition can be extremely computationally demanding. The primate visual system is not isolated from other perceptual and motor systems; therefore, it is likely that a recognition system would utilize all available evidence, whatever the input modality, to accomplish its task. Studying object recognition using a single input modality overlooks the importance of the integration of inputs. To explore the interplay between visual and haptic inputs for object recognition, we used behavioral, neuropsychological, and neuroimaging techniques. The data from these studies converge to suggest that shared mechanisms for volumetric shape processing across vision and touch are found in the lateral occipital complex and the anterior intraparietal sulcus. Processing of visual-haptic shape in two separate locations suggests that the two visual streams theory may extend to dorsal and ventral streams of visual-haptic shape integration for action and perception.

2/18:    Jun Zhang, Department of Psychology, University of Michigan, Ann Arbor
Unconfounding Stimulus - and Response - Related Processes and Determining Sensory-Motor Locus in Neural Recordings

When a neural signal (single neuron activity, ERP waveform, fMRI activation, etc) is being recorded simultaneously as the animal/human subject performs a behavioral task, a question arises as to whether the recorded neural signal reflects encoding/analysis of the stimulus, preparation/execution of the response, or the "decision" that translates a stimulus into a response. I will present several techniques that allow us to 1) unambiguously determine the "locus" along sensory-motor arc of the recorded neural activity based on their trial-by-trial correlation with behavior; 2) uniquely recover a stimulus-locked and a response-locked component in the recorded neural activity based on trial-by-trial variability in response time (RT). Applying 1) to single neuron recording data allows a refined analysis and characterization of LIP accumulation neurons in a random-dot motion discrimination task. Applying 2) to ERP data clarifies a long-standing debate in ERP literature concerning the so-called "P3 anteriorization" phenomenon for Go/Nogo tasks.

2/25:    Joe Houpt
Measuring Visual Word Processing Efficiency

Visual word perception is a fundamental part of reading and as such has been the focus of much attention in cognitive psychology. Many of the most influential studies of word perception have focused on the efficiency of word perception, but have used different baselines of comparison. This inconsistency has led to diverging models of visual word perception. In this paper we use the workload capacity coefficient measure of efficiency to address this inconsistency. This measure has the advantage of using the predicted performance of a standard processing model as a baseline. Based on these analyses, we find evidence higher efficiency in word processing than the baseline model, as well as better than non-words and upside-down non-words. Some participants showed increased efficiency for pseudoword processing, although the effect was not as regular as the word processing efficiency.

3/4:    Dan Fogerty
Age Differences in Processing Auditory Temporal Sequences

Many older listeners have reduced speech understanding abilities that persist even after audibility of the signal is restored. Declining abilities in temporal processing have been proposed as one source of this underlying difficulty as adults age. This study examined the temporal-order processing abilities of young and older adults for several different auditory vowel sequence conditions. Results indicated that older listeners had more variability and performed poorer than young listeners on vowel-identification tasks, although a large overlap in distributions was observed. In addition, older listeners improved performance with additional stimulus exposure, but did not match the performance of young listeners. Individual differences among older listeners demonstrated the influence of cognitive factors, but not audibility or age.

3/11:    Skyler Place
Utilizing Social Information During Mate Choice

When searching for a mate, one must gather information to determine the mate value of potential partners. By focusing on individuals that have been previously chosen by others, one’s selection of mates can be influenced by another’s successful search – a phenomenon known as mate copying. We show evidence of mate copying in humans with a novel methodology that closely mimics behavioral studies with non-human animals and goes beyond the use of staged still-picture stimuli in previous human mate copying studies. After viewing instances of real mating interest in video recordings of speed-dates, both male and female participants demonstrate mate copying effects for short-term and long-term relationship interest when they perceived the dates as successful. This first study is followed by a second that illustrates when and how mate copying can be used in conjunction with other mate value cues.

3/25:    Cameron Buckner
How I Learned to Stop Worrying about Mental Causation and Love the Hippocampus: A Neo-Dretskean Account of Reasons as Causes

Intentional explanations—explanations which account for behaviors in terms of mental states with semantic contents, like beliefs and desires—are frequently deployed in psychology and cognitive ethology. Computationalism purports to legitimize these explanatory appeals to semantic content by suggesting that minds are syntactic, information-processing engines. Minds are therefore declared to be possible in a physical world because mental states have formal, nonsemantic properties which correspond to their semantic properties, and thus mental states can be manipulated to produce behavior using rules sensitive only to their syntax. A straightforward problem with this approach to intentional explanation, however, is that semantic properties themselves are apparently not causally relevant to the explanation of behavior, with syntax doing all the real causal work. In the 1980s, an alternative approach to intentional explanation emerged called teleosemantics. Teleosemanticists held that semantic properties are causally relevant to the explanation of behavior in their own right, because the semantic properties of mental states caused them to be selected for positions of behavioral control in the organism’s history (either through evolution or learning). By the mid-to-late 1990s, popular opinion in philosophy of mind held that the approach had run into serious problems, and work in this area began to wane. In this talk, I will argue that there is life left in the teleosemantic program yet, especially given that proponents of the view never adequately made touch with the relevant psychology and neuroscience (which has progressed in leaps and bounds since the position was first developed). By drawing upon current cognitive neuroscience of the hippocampus, I will argue that if much of this work is right in its broad strokes, then many of the most common objections to teleosemantics can be rebutted with straightforward appeals to widely-available empirical findings. My hope is that the resulting view will provide a fully naturalizable theory of content which is of real use to the cognitive sciences.

4/1:    John K. Kruschke
An Introduction to Bayesian Data Analysis

Bayesian methods have garnered huge interest in cognitive science as an approach to models of cognition and perception. On the other hand, Bayesian methods for data analysis have not yet made much headway in cognitive science against the institutionalized inertia of 20th century null hypothesis significance testing (NHST). Ironically, specific Bayesian models of cognition and perception may not long endure the ravages of empirical verification, but generic Bayesian methods for data analysis should eventually dominate. It is time that Bayesian data analysis became the norm for empirical methods in cognitive science. This talk reviews a fatal flaw of NHST and introduces the audience to some benefits of Bayesian data analysis. The talk presents illustrative examples of multiple comparisons in Bayesian ANOVA and Bayesian approaches to statistical power.

4/8:    Stephen E. Denton and John K. Kruschke
A Bayesian Framework for Active Learning

Bayesian approaches to learning involve incrementally updating degrees of belief across a space of hypotheses whenever an observer passively observes a stimulus and outcome. But these approaches also provide a framework for models of active learning, because uncertainty across beliefs can be evaluated and thus the expected uncertainty reduction for candidate stimuli can be computed. Bayesian models of active learning predict that an active learner would select the stimulus for which expected uncertainty across all hypotheses is minimized. Our research contrasts four possible hypotheses spaces consisting of two simple cue-combination models and two possible priors. To tease apart the models, we performed an automated search of associative learning structures for which the models make maximally different predictions. Human participants were tested on these same learning structures in the context of an allergy diagnosis task. At various points in training we asked which cues they would find most informative to learn about; i.e., we assessed the participant's active learning preferences. The model and prior combinations that best mimic human active learning will be discussed.

4/15:    Leslie Blaha
An LDA Approach to the Neural Correlates of Configural Learning

The purpose of my current study is to employ linear discriminant analysis (LDA; Philiastides & Sajda, 2006) to characterize the changes in ERPs over the entire course of a perceptual learning task. Configural learning is the perceptual learning process by which participants develop configural processing strategies or representations characterized by extremely efficient parallel information processing (Blaha & Townsend, under revision). Participants performed a perceptual unitization task in which they learned to categorize novel images. Correct categorization responses required exhaustive feature identification, which encouraged unitization of images into unified object percepts. Linear discriminator accuracy, measured by Az, increased each day of training, showing significant differences in neural signals between categories on and after training day 3 or 4 for all participants. Additionally, the LDA training window starting time resulting in discriminator performance of 65% accuracy or better shifted from 450-500ms to 300ms after stimulus onset at the completion of training. LDA results are consistent with our earlier report (Blaha & Busey, VSS 2007) of peak ERP amplitude differences between categories after training at approximately 170ms and 250ms after stimulus onset. These EEG results are consistent with the hypothesis that perceptual unitization results in configural perceptual processing mechanisms.

4/22:    Paul Williams
The Dynamics of Information Flow in Embodied Relational Categorization

Information theory provides a powerful set of tools and concepts for the analysis of embodied cognitive systems. To date, such analyses have typically ignored the temporal behavior of such systems, instead collapsing over time to apply static measures of information structure. However, as I will discuss, the same basic concepts of information theory can also be used to explore the behavior of a system as it unfolds through time, giving rise to a notion of information flow. In this talk, I will present results from an information-theoretic analysis of a model agent that performs a simple kind of relational categorization. I will demonstrate how, using techniques to characterize the structure of information flow, we can rigorously formulate and address questions such as how the agent extracts and stores information about stimulus features, and how it integrates information about multiple features. Finally, I will argue that when applied in this way information-theoretic techniques are in fact closely related to those of dynamical systems theory, and provide a complementary picture of how the behavior of a system emerges through the specific interactions of its components.

4/29:    Douglas Hofstadter
Why Translation is Not Mechanical and Why Translators Cannot be Invisible

I will discuss the nature of translation, showing why artistic decisions are involved left and right in every act of translation. Although translators often say that their goal is to be "invisible servants" of their author, this is a vain and naïve hope. I'll use some hopefully entertaining examples to demonstrate why this is the case. (Incidentally, Professor Robert Goldstone has informed me [Personal communication, 24 April 2009] that this talk would qualify as a "cognitive lunch" since -- so he assures me -- translation is a highly cognitive phenomenon.)

5/6:    Benjamin Scheibehenne
Can there ever be too many flowers blooming? Re-visiting the effect of choice overload

The choice overload- or too-much-choice effect states that having too many options to choose from may lead to negative consequences such as a decreases the motivation to choose and/or a lower satisfaction with the finally chosen option. A number of studies in the past found strong instances of this effect in the lab and in the field that raised a lot of attention in the scientific community as well as in the media. In my talk I will lay out that the theoretical explanations of the effect are somewhat sparse. I will further present a series of experiments that could not replicate the effect. Based on a meta-analysis on published and unpublished experiments on choice overload I will then show that its mean effect size is zero with hardly any effective moderators. Together these results seem to suggest that choice overload is not a reliable psychological phenomenon. However, I will also review some moderators that might still explain the diverging results between those studies that found the result and those that did not, some of which might be worth testing in the future.