Learning emergent behaviours for a hierarchical Bayesian framework for active robotic perception

In this research work, we contribute with a behaviour learning process for a hierarchical Bayesian framework for multimodal active perception, devised to be emergent, scalable and adaptive. This framework is composed by models built upon a common spatial configuration for encoding perception and action that is naturally fitting for the integration of readings from multiple sensors, using a Bayesian approach devised in previous work. The proposed learning process is shown to reproduce goal-dependent human-like active perception behaviours by learning model parameters (referred to as "attentional sets") for different free-viewing and active search tasks. Learning was performed by presenting several 3D audiovisual virtual scenarios using a head-mounted display, while logging the spatial distribution of fixations of the subject (in 2D, on left and right images, and in 3D space), data which are consequently used as the training set for the framework. As a consequence, the hierarchical Bayesian framework adequately implements high-level behaviour resulting from low-level interaction of simpler building blocks by using the attentional sets learned for each task, and is able to change these attentional sets "on the fly," allowing the implementation of goal-dependent behaviours (i.e., top-down influences).     

Up north and down south: Implicit associations between topography and cardinal direction

Route planners show a reliable tendency to select south- relative to north-going routes between two horizontally (east/west) aligned landmarks, suggesting the application of a north-is-up heuristic (Brunyé, Mahoney, Gardony, & Taylor, 2010). The source of this north-is-up bias remains unknown, and there is no strong evidence to suggest that it is due to explicit strategy use. In four experiments, we attempt to further elucidate the source of this effect by testing whether it can be attributed to implicit associations between cardinal direction (north/south) and topography (mountainous/level terrain). Experiments 1 and 2 used an adapted Implicit Association Test and demonstrate automatically activated judgements that associate north with mountainous and south with relatively level terrain. Experiment 3 rules out the possibility that this effect is due to the local topography of New England by replicating in participants from the topographically dissimilar Midwestern United States. Finally, Experiment 4 tests the relative contribution of implicit versus explicit associations between cardinal direction and topography in predicting route-planning asymmetries; we show that implicit associations are a stronger predictor of southern route biases than explicit processes. Overall, results demonstrate that the conceptualization of space can be driven by physically unfounded implicit associations between cardinal directions and topographical features, and these associations are at least partially responsible for southern route preferences.     

Characteristic patterns of word association responses in institutionalized elderly with and without senile dementia

This study examined the responses of 91 institutionalized elderly persons to the Goldfarb-Halpern Word Association Test (1981), Journal of Speech and Hearing Research, 24, 233–246. Subjects with senile dementia evidenced a characteristic pattern of responses which included marked reduction of paradigmatic responses, no decrease in syntagmatic responses, and a marked increase in unclassifiable and multiword responses not previously reported in the literature. The declines and increases appeared linear and progressive with severity of dementia. The pattern was distinct from that of any other adult group previously tested with this tool. Institutionalized elderly subjects without senile dementia performed similarly to previously tested noninstitutionalized elderly. Theoretical and clinical implications are discussed.     

Negotiating transitions: a supervision model for home-based family therapists

Home-based family-centered treatment programs provide mental health and case management services to families in their own home. Treatment objectives include the prevention of residential or psychiatric placement, strengthening the family unit, and promoting family self-sufficiency. Little has been written, however, regarding the supervision of home-based, family-centered therapists. The purpose of this article is to describe a supervision model to teach novice-level family therapists to think systemically and to do family therapy with multiproblem at-risk families.     

Analogical transfer in young children: Analogies as tools for communication and exposition

Children's ability to transfer what they learn in one situation to analogous problems was examined in a series of four studies. Subjects ranged in age from 3 to 10 years. The problems involved novel uses of familiar tools or simple biological themes such as mimicry as a method of defence. The data suggest that the apparent transfer reluctance shown by children in previous studies is the result of what they have been required to learn and the conditions under which they have been required to learn it. In the present studies, children as young as 3 years transferred readily if the problem domain was one they understood and engaged in, and if the traditional laboratory paradigm was modified so as to promote transfer rather than just to test for its spontaneous occurrence.     

Use of a raster framebuffer in vision research

For the study of human and animal vision, the video framebuffer is the only technology that is capable of displaying two-dimensional images with precise control of contrast, luminance, and display timing. The video framebuffer also allows precise control of color. However, this device is not designed for precise psychophysical displays, and techniques must be developed to use them in this role. In this paper, we describe a number of basic methods used in our lab. In order to be concrete, we use an Adage RDS-3000 raster display system (Adage, 1982) as an example, since that is the device we have actually used. The Adage is hosted by a PDP-11/73 under the Venix operating system. The principles generalize to other machines. Where it clarifies the issues, we have shown the syntax of the software routines involved.     

Aging and memory for faces versus single views of faces

Differences in the ability of young adults and elderly to recognize faces were examined under two conditions. In a standard single-view condition, in which each input face was shown as one photograph, we confirmed prior findings that young adults perform better than the elderly at distinguishing photographs seen before from photographs of new faces. We also found that the elderly had more trouble distinguishing photographs seen before from photographs of(l) old faces changed in facial expression and (2) old faces changed in expression and pose. Yet there were no reliable age differences in distinguishing old-but-changed faces from entirely new faces. In a more naturalistic multi-view condition, in which each input face was shown in four poses and with two expressions, no age differences were found. A second experiment ruled out the possibility that varied repetition, by itself, removes age differences in recognizing faces. These data supported age differences in remembering facial expressions and possibly other details of photographs of faces, but not in remembering faces perse.     

On time differences in searching for letters in words and nonwords: Do they emerge during the initial encoding or the subsequent scan?

Krueger (1970a, 1970b, 1982) has demonstrated that subjects can search for target letters within words faster than they can complete an equivalent search through nonwords, and he further demonstrated that the effect did not arise during the comparison stage. The present study involved three experiments in which the usual word advantage disappeared either when subjects knew where within a display the target item would appear (i.e., it was always the first letter), or when all the component letters were encoded into memory before the task began (i.e., a memory-search task). These data, in conjunction with Krueger's, where interpreted as localizing at least one (and possibly the only) source of the word-nonword difference in this task to the events that occur during the item-to-item transitions subjects make when scanning the letter arrays. That is, these transitions are faster for words than nonwords, and it was suggested that the time difference may emerge because although all the letters from within a word appear to be available in memory before the scan begins, this seems not to be true for consonant arrays. Given that this is the case, part of the word-nonword difference may be attributable to subsequent encoding events that would be needed for the consonant arrays as the scan moves from letter to letter.