|
The Cognitive Lunch talks will be on Wednesdays from 12:10 - 1:25 in the Psychology conference room (PY 128) located behind the main office.
Abstract 9/6: Richard M. Shiffrin
Rational Inference
I use a variant of the 'Exchange Paradox' to motivate a discussion of the psychological basis of rationality. I argue that the dependence of rationality upon the local context and a social consensus puts limits upon the reach and application of rational inference. To provide an example, I use a variant of the exchange paradox to show that a common and almost universal principle, often used in Bayesian inference, but also in almost all forms of inference, can lead to irrational decision making. That principle states: Having observed some data to be used as a basis for inference and action, one should limit consideration to scenarios that are consistent with that data, and ignore what data could have been observed, but wasn’t. To whet the audience appetite, the paradox follows: Suppose the game organizer flips a coin until a heads appears on flip n, and places 10**(n) and 10**(n+1) dollars in each of two sealed envelopes. With probability 0.8 the decision maker is handed the envelope with the larger amount (else the envelope with the smaller amount). The decision maker opens the envelope and observes X, and either decides to keep that value, or irrevocably trade it for the value in the other (which will be either X/10 or 10X). The goal is to maximize expected value. Strangely, it can be shown that use of the above mentioned principle will lead the decision maker to exchange regardless of X. It seems paradoxical to 'always' exchange what you know to be the envelope with the higher probability of having the larger amount. This is sufficiently weird to lead most thinkers to decide that this decision and action is irrational. If so, we have a demonstration that the principle has limited applicability. 9/13: Matt Jones, U. of Texas at Austin
Sequential Effects in Categorization: Implications for Category
Representation, Attention, and Similarity-based Generalization
Sequential effects are pervasive in any repeated cognitive task, yet are
generally neglected and treated as noise. Here, I show how
consideration of these effects in category learning provides a powerful
tool for addressing questions of category representation, selective
attention, and the role of similarity in generalization. 1.The effect of learning on one trial upon the response on the next
supports exemplar-generalization theory over theories based on
prototypes, decision bounds, or rules. 2. This pattern of sequential effects can be used to directly measure
subjects' generalization gradients during category learning. 3. Comparison of the generalization gradient for different category
structures allows testing of attentional learning models. 4. Results from a 4-category task shows that generalization cannot be
mediated by similarity, as it depends on the full multidimensional
relationship between stimuli. Implications will also be discussed for the roles of short- and
long-term memory in category learning. 9/20: Michael Jones
TBA
TBA9/27: Adam Sanborn
TBA
TBA10/4: David Gilden
TBA
TBA10/11: Evelina Dineva
Dynamical Field Theory of Infant Perseveration: The Stabilization of Decisions allows for a Robotic Enactment.
After introducing Piaget's A-not-B paradigm in which infants show perseverate reaching under some circumstances, I will present the dynamical field theory (DFT) approach, which postulates that A-not-B is a decision making task, and that young infant's reaching decisions are input driven. Part of this input is an infant's own behavioral history (preshape trace) that is dynamically generated with each reach.Focus of the talk will be that stabilization---temporal persistence---of a reaching decision is a prerequisite for the formation of behavioral history. This is because a dynamically stable motor plan is necessary for constant behavior. Thus, stabilization is also necessary the embodiment of the DFT model on a robot. In the last part of the talk I will present how we implement the DFT model as a motor planning system on a robot, and will show how the robot behaves in several A-not-B situations. 10/18: Thomas Hills
TBA
TBA10/25: Rick Hullinger
Attention To Individuating Cues Obviates the Need for Dual Process Theories
van Osselaer, Janiszewski, and Cunha (2004) suggested that people learn by using both an exemplar-based associative learning process and a second, component-cue based learning process. We present an alternative explanation of their results and show that a critical detail of their experiment was overlooked. We will present data from four new experiments that isolate the effect of the missing detail and provide a better understanding of the learning process. When the critical detail is incorporated into the analysis, we will show that the results can be explained by a single exemplar-based learning process that attends strongly to individuating cues.11/1: Ami Eidels
When to sell your used car? The things that hazard functions can
tell us
In reliability theory, the hazard function, also called the failure
rate, tells us the probability that a machine (or a component) will
fail at time t, given that it has not failed. While it can
theoretically tell us when is the best time to sell our used car, its
benefits to cognitive psychology are more evident: probability
distributions with similar density and cumulative distribution
functions can have different hazard functions, making the latter useful
in distinguishing between models and understanding the characteristics
of the mental process under investigation.In the current study we derived and simulated hazard functions of
different processing models (namely, the two-state varied model, three
types of diffusion models, the race model) and showed that they may
predict different shapes of the hazard function. We proposed (and
simulated) an even more fine-grained analysis by introducing the
conditional hazard functions: analyzing the hazard functions separately
for correct and incorrect responses. Finally, we collected data from a
simple two alternative forced choice task and compared the estimated
hazard functions, standard and conditional, to those predicted by the
different models. 11/8: Ji Son
TBA
TBA11/15: Winter Mason
TBA
TBA11/29: Alfredo Pereira
TBA
TBA12/6: Chen Yu
Multimodal Statistical Learning: Linking Words to World
There are an infinite number of possible word-to-world pairings in naturalistic learning environments. Previous proposals to solve this mapping problem focus on linguistic, social, and representational constraints at a single moment. We examined an alternative account -- a cross-situational learning strategy based on computing distributional statistics across words, across referents, and most importantly across the co-occurrences of these two at multiple moments. The results from a series of experiments showed that cross-situational statistical learning is within the repertoire of human learners. We also proposed and implemented a set of computational models and feed them with the same training data used in different learning conditions in experimental studies, to shed light on the possible underlying mechanisms of statistical learning. Moreover, we suggest that social cues, as embodied multimodal interactions between language teachers and language learners, can be integrated in the cross-situational statistical framework.I will briefly present several multisensory experimental setups and results. |
