Efficient Learning of Relational Object Class Models

We present an efficient method for learning part-based object class models from unsegmented images represented as sets of salient features. A model includes parts’ appearance, as well as location and scale relations between parts. The object class is generatively modeled using a simple Bayesian network with a central hidden node containing location and scale information, and nodes describing object parts. The model’s parameters, however, are optimized to reduce a loss function of the training error, as in discriminative methods. We show how boosting techniques can be extended to optimize the relational model proposed, with complexity linear in the number of parts and the number of features per image. This efficiency allows our method to learn relational models with many parts and features. The method has an advantage over purely generative and purely discriminative approaches for learning from sets of salient features, since generative method often use a small number of parts and features, while discriminative methods tend to ignore geometrical relations between parts. Experimental results are described, using some bench-mark data sets and three sets of newly collected data, showing the relative merits of our method in recognition and localization tasks.     

When do base rates affect predictions?

M. Manis et al (see record 1981-07963-001) found that base rates had a clear effect on discrete predictions and a smaller effect on the confidence that Ss attached to those predictions. As a result, the findings of Manis et al can be reinterpreted in a way that makes them compatible with previous findings. In this light, their study emerges as a constructive replication of earlier results demonstrating judgment by representativeness. (15 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Why are people reluctant to exchange lottery tickets?

Students were given lottery tickets and then were asked to exchange their ticket for another one, plus a small monetary incentive. Less than 50% agreed. In contrast, when given pens, and the same exchange offer, over 90% agreed. Experimental control rules out that the reluctance to exchange lottery tickets results from (1) overestimation of the winning probability of one's own ticket; (2) a concern that the ticket, once exchanged, might win in the hands of another; (3) from an overly low transaction cost; (4) "paranoia" due to the bizarreness of the social situation we created; or (5) "bonding" to the ticket. Rather, the possibility for ex post regret that exists when exchanging lottery tickets, but not pens, underlies this reluctance. The notion of regret is broader than previously supposed. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

Misconceptions in Martin and Terris's (1991) "Predicting infrequent behavior: Clarifying the impact on false-positive rates."

S. L. Martin and W. Terris (see record 1991-28965-001) recently attributed to a number of psychologists a concept they called the false-positive argument (FPA), according to which a test should not be used if an individual who fails is more likely to be qualified than unqualified, and they attempted to clarify the conditions under which the FPA may be appropriate. It is argued that in none of the articles cited by Martin and Terris is the FPA truly posited and also that they failed to clarify the conditions under which the FPA might be appropriate. These conditions depend on the costs and payoffs associated with the various outcomes of the decision problem, as is demonstrated through the use of a threshold utility model. Finally, the examples used by Martin and Terris deal with the detection of deception but were discussed without proper consideration of the contexts in which lie-detection techniques are typically used. (PsycINFO Database Record (c) 2010 APA, all rights reserved)     

An evaluation of polygraphers' judgments: A review from a decision theoretic perspective.

Reexamined data from 7 field studies to estimate the discriminability of the control questions technique (CQT) in real-life situations. A signal detection model was applied, and an attempt was made to derive the value system of the polygraphers who participated. It was demonstrated that under an assumption of rationality, the examiners tended to value the detection of guilty suspects highly, even in the presence of a high risk of falsely classifying innocent suspects as deceptive. An index of usefulness of the CQT-based evaluation system was defined, and the CQT's range of usefulness was examined as a function of operating costs, possible payoff ratios, and prior probabilities of guilt. It is unlikely that the CQT method of lie detection would be useful for determining guilt in a court of law or even for preemployment screening, although it might be useful for police investigative procedures. (30 ref) (PsycINFO Database Record (c) 2010 APA, all rights reserved)