• 01/17   Chris Rafael, Indiana University School of Informatics - abstract
• 01/25   Jim Townsend, Indiana University
• 02/01   Ehtibar Dzhafarov
• 02/08   William Alexander
• 02/15   Young Lee - abstract
• 02/22   Brian Stankiewicz
• 03/01   Stephen Denton - abstract
• 03/08   Eric Dimperio - abstract
• 03/22   Aaron Benjamin - abstract
• 03/29   Kelly Addis, Indiana University abstract
• 04/05   David Landy
• 04/12   Krystal Klein
• 04/19   Josh Goldberg, Indiana University abstract
• 04/26   Anthony Bishana

01/17: Chris Rafael
Music Plus One

I will discuss my ongoing work in creating a computer system that simulates a sensitively conducted orchestra in a non-improvisatory composition for soloist and orchestra.

My accompaniment system synthesizes a number of knowledge sources including the musical score, on-line analysis of the soloist's performance, and the musical interpretations demonstrated by both the soloist and orchestra in rehearsal. I present a probabilistic model --- a Bayesian Belief Network that represents these disparate knowledge sources in a coherent framework.

During live performance, my system "listens" to the soloist by using a hidden Markov model and conducts the orchestra through principled real-time decision-making engine that incorporates all currently available information for each decision. I will provide a live demonstration of my system on an excerpt from the Strauss Oboe Concerto.

Many judgment studies have found that visual perception of object distance, size, and shape are both imprecise and inaccurate. Given the possibility that there are two visual systems and that judgments tap one system but the other exists to guide actions, previous studies tested whether action measures might reveal good perception of distance, size and shape. Studies of feedforward reaching failed to confirm this and showed that space perception was equally poor as found in previous judgment studies. However, in the context of visually guided actions feedback might be used to calibrate space perception. Studies of visually guided reaches-to-grasp have shown that indeed the visual perception of object distance and size are each calibrated by feedback to yield accurate and precise feedfoward reaches-to-grasp.

We now investigate whether shape perception can be calibrated in the context of reaches-to-grasp. One aspect of shape is the aspect ratio of depth (extent along the gaze axis) to width (extent in a frontoparallel plane). These ratios are typically perceived inaccurately. We used cylindrical objects with elliptical cross section of varying eccentricity (e.g. flat to round to elongated). Our participants reached to grasp the width or the depth of these objects with their index finger and thumb. The maximum grasp aperture (MGA) and the terminal grasp aperture (TGA) were used to evaluate perception. MGA occurs at 50-70% of a reach, and TGA occurs at the end of a reach but before the hand has contacted an object. In Experiments 1 and 2, we used the same approach as used in previous distance and size perception studies and investigated whether perceived shape is recalibrated by distorted haptic feedback. The results showed that it is not! In Experiments 3 and 4, we tested the accuracy of feedforward (or open loop) grasping with respect to shape. The results showed that both TGA depth/width and MGA depth/ width were too large for small aspect ratios and too small for large aspect ratios. Feedforward grasping reflected poor shape perception just as in previous judgment studies. In Experiment 5, we hypothesized that on-line guidance is needed for accurate grasping. In condition 1, participants reached to grasp with vision of the hand (closed loop), while in condition 2, they reached to grasp without vision of hand (open loop). The result was that the former was relatively accurate and precise while the latter was not. We concluded that shape perception cannot be calibrated by feedback from reaches-to-grasp and that on-line visual guidance is required for accurate grasping because shape perception is poor.

The associative learning effect called blocking has previously been found inmany cue-competition paradigms where all cues are of equal salience. Previous research by Hall, Mackintosh, Goodall, and Dal Martello (1977) found that, in animals, salient cues are less likely to be blocked. Crucially, they also found that when the to-be-blocked cue is highly salient, the blocking cue will lose some control over responding. I will present research that extends these findings to humans and suggests that shifts in attention can explain the apparent loss of control by the previously learned cue. A connectionist model that implements attentional learning is shown to fit the main trends in the data. Model comparisons suggest that mere forgetting, implemented as weight decay, cannot explain the results.

Function learning tasks require participants to learn relationships between continuously variable cue values and continuously variable response values where the relationship is defined by a mathematical function. Both the Extrapolation-Association Model (EXAM) and the Population of Linear Experts (POLE) model seek to explain how people extrapolate a learned function when presented with novel cue stimuli. A comparison of extrapolation behavior was performed by looking at each model’s ability to match data of individual subjects. Both models had there strengths and weaknesses, but both failed to capture certain patterns seen in the data.

Empirical dissociations in recognition memory have been claimed as a strong basis for postulating theoretical dissociations in memory systems or processes. This talk will challenge that approach by examining the behavior of simple models that do not have dissociable memory capacities. These models reveal that even ordinal dissociations can arise under realistic assumptions about the nature of memory representations. Empirical tests of this perspective will be presented.

Much attention has been given to the von Restorff (or isolation) effect. If a single stimulus is differentiated from all others in a study list by means of physical, perceptual or semantic properties, this isolated item shows an increased recall probability relative to a control item. It is reasonable to assume that several list items must be presented before the isolate item can be correctly identified. In fact, this assumption is so pervasive that many studies fix the isolate at a single serial position late in the list. However, this assumption has not been rigorously tested, particularly in the case of semantic isolation where differentiation may rely on deeper stimulus processing. In the present study, we manipulate the serial position of a semantic isolate item in both immediate and delayed free recall paradigms. The results indicate that the isolate must occur in a late serial position to produce the von Restorff effect. Additionally, isolate items at early serial positions produce a decrease in recall probability. Implications for current theories will be discussed.

A feature that is left out of models of infant habituation is the role that looking behaviors themselves play in determining the time structure of what an infant sees. Since inferences from looking times form the foundation of our understanding of infant cognition, it is important to understand this embodied nature of looking and what that means for cognition.

I will present a model I am developing with Gregor Schöner, following up on his habituation model with Esther Thelen (Schöner & Thelen, In Press) and also heavily based on insights from the dynamical field model of perseverative reaching (A-not-B). (Thelen et al, 2001) The model is a work in progress, but I will explain its motivations, and then step through its operations to show how it accounts for the balance between stabilizing fixation while allowing for shifts of gaze, as well as the transition from familiarity to novelty preferences as measured in its looking times. In the near future, we expect to be able to shed much-needed light on the differences in apparent knowledge that babies exhibit between habituation and Visual Paired Comparison task designs.

Although habituation experiments sometimes do overreach, (see, e.g., a special issue of Infancy (2000) vol. 1, no. 4) the intuition that looking behavior gives us a window into cognitive processes like memory, learning, and attention is a good one. This model aims at a much more unified understanding of what underlies these behaviors.

Reference:
Schöner, G. and Thelen, E. (In press). Using dynamic field theory to rethink infant habituation. Psychological Review.

Thelen, E., Schöner, G., Scheier, C., and Smith, L. B. (2001). The dynamics of embodiment: A field theory of infant perseverative reaching. Behavioral and Brain Sciences, 24:1–86.