Search:  Hint: Leave blank to view All Records Linda B. Smith , Chancellors Professor     Organization: Indiana University Primary Field: Psychology Secondary Field: About: Cognitive development and in particular interactions among language, perception, and action. jonathan d. nelson , Dr     Organization: University of California, San Diego Primary Field: Cognitive Science Secondary Field: Artificial Intelligence About: C. Teuscher and J. Triesch. To Each His Own: The Caregiver's Role in a Computational Model of Gaze Following o Each His Own: The Caregiver's Role in. .     Abstract: We investigate a computational model of the emergence of gaze following that is based on a generic basic set of mechanisms. Whereas much attention has been focused so far on the study of the infant's behavior, we systematically analyze the caregiver and show that he plays a crucial role in the development of gaze following in our model, especially for infant models with simulated developmental disorders such as autism and Williams syndrome. We first create two reference infant parameter sets and test their behavior with a simple standard caregiver. Based on these findings we then propose new caregiver models and evaluate them on normally developing infants and on infants with simulated developmental disorders. Further, we investigate if and how a pair of infants (with and without simulated developmental disorders) might learn gaze following from scratch, without a mature caregiver. The findings of this paper suggest the pivotal role the caregiver plays for the infant in developing gaze following, that his predictability is the most important criterion, and that different infant models require particular caregivers for gaze following to emerge optimally. Our simulations are consistent with Leekam's finding, that autistics can learn to follow gaze through a contingent presentation of rewarding visual stimuli, but that a lack of motivation may impede learning. Field: Computer Vision Fumihide Tanaka, Aaron Cicourel, and Javier R. Movellan. Socialization between toddlers and robots at an early childhood education center. .     Abstract: A state-of-the-art social robot was immersed in a classroom of toddlers for >5 months. The quality of the interaction between children and robots improved steadily for 27 sessions, quickly deteriorated for 15 sessions when the robot was reprogrammed to behave in a predictable manner, and improved in the last three sessions when the robot displayed again its full behavioral repertoire. Initially, the children treated the robot very differently than the way they treated each other. By the last sessions, 5 months later, they treated the robot as a peer rather than as a toy. Results indicate that current robot technology is surprisingly close to achieving autonomous bonding and socialization with human toddlers for sustained periods of time and that it could have great potential in educational settings assisting teachers and enriching the classroom environment. Field: Artificial Intelligence Thomas R. Shultz et al.. Knowledge-based Learning with KBCC. .     Abstract: Abstract – A constructive learning algorithm, knowledgebased cascade-correlation (KBCC), recruits previously-learned networks in addition to the single hidden units recruited by ordinary cascade-correlation. This enables learning by analogy when adequate prior knowledge is available, learning by induction from examples when there is no relevant prior knowledge, and various combinations of analogy and induction. A review of experiments with KBCC indicates that recruitment of relevant existing knowledge typically speeds learning and sometimes enables learning of otherwise impossible problems. Current limitations of this approach are discussed. Field: Nicholas J. Butko et al.. Learning about Humans During the First 6 Minutes of Life. .     Abstract: There is strong experimental evidence that newborn infants orient towards human faces [1]. While opinions are divided as to whether this preference reflects domain specific knowledge about the appearance of human beings, or general preferences for stimuli that happen to occur in humans [2] most views agree that the face-preference phenomenon is innate and not learned. Here we explore another hypothesis, the Rapid Learning Hypothesis which in the past was rejected as being computationally implausible. We built a robot with the appearance of a human baby and endowed it with an algorithm that detects contingencies in the auditory domain [3]. Members of our laboratory were then encouraged to sporadically interact with the baby robot. We also endowed the robot with a new machine learning algorithm for discovering visual concepts [4]. The input to the system were the images collected by the baby robot’s camera at 30 frames per second. In addition each image was automatically labeled by the auditory contingency detector to indicate whether or not auditory contingencies were present at the time the video frame was captured. 30% of the images captured while auditory contingencies were detected, did not contain people. 5% of the images captured while auditory contingencies were not detected, contained people. In addition, people could appear anywhere on the image plane, sometimes showing their face, sometimes other parts of their body. In less than 6 minutes of interaction with the world the robot learned to locate people in novel images. In addition, it developed a preference for drawings of human faces over drawings of non-faces, even though it had never been exposed to such schematic face drawings before. During learning, the baby robot was never told whether or not people were present in the images, or whether people were of any particular relevance at all. It simply discovered that to make sense of the images and sounds it received, it was a good idea to use feature detectors that happened to discriminate the presence or absence of people. While in our experiment we used auditory contingency as a training signal for a visual concept learner, other training signals could also have been used. All that is required is for the signal to provide higher than chance information about the presence or absence of people. For example, if a baby or robot is being touched or moved, these could likely serve also as a training signal. The results illustrate that visual preferences of the type typically investigated in human neonates can be acquired very quickly, in a matter of minutes. Previous studies that were thought to provide evidence for innate cognitive modules may actually be evidence for rapid learning mechanisms in a neonate brain exquisitely tuned to detect the statistical structure of the world. Field: John K. Kruschke. Learned Attention. .     Abstract: Unlike many approaches to machine learning, human learning involves selective attention. When confronted by new things to learn, people can rapidly shift attention, thereby increasing speed of acquisition and decreasing interference with previous knowledge. The shift of attention is itself learned, so that attention is allocated to particular cues in particular contexts. While selective attention benefits acquisition, it can also lead to distortions of knowledge that are evident when the knowledge is transferred to novel situations. Several mathematical models have been designed to implement selective attention in learning; the models quantitatively fit human performance in many experiments. This presentation reviews various research projects of the author Field: Li Fei-Fei. Knowledge transfer in learning to recognize visual objects classes. .     Abstract: Learning to recognize of object classes is one of the most important functionalities of vision. It is estimated that humans are able to learn tens of thousands of visual categories in their life. Given the photometric and geometric variabilities displayed by objects as well as the high degree of intra-class variabilities, we hypothesize that humans achieve such a feat by using knowledge and information cumulated throughout the learning process. In recent years, a handful of pioneering papers have applied various forms of knowledge transfer algorithms to the problem of learning object classes. We first review some of these papers by loosely grouping them into three categories: transfer through prior parameters, transfer through shared features or parts, and transfer through contextual information. In the second half of the paper, we detail a recent algorithm proposed by the author. This incremental learning scheme uses information from object classes previously learned in the form of prior models to train a new object class model. Training images can be presented in an incremental way. We present experimental results tested with this model on a large number of object categories. Field: Susan S. Jones. Infants Learn to Imitate by Being Imitated. .     Abstract: An account is outlined in which the development of imitation in human infants relies on associative learning rather than on innate knowledge. Initial studies that support this learning account are reported. Field: Nigel Singh and Emanuel Todorov. Imitation Learning for Reaching in Virtual Reality Environments. .     Abstract: Results for an imitation learning reaching controller for a Virtual Reality environment with simulated physics is presented. The controller follows the unforced dynamics of a localized vector field around the human example data. Trajectory and velocity profiles compare well with human data. Future experiments with more complex tasks are discussed. Field: Page 1 of 44 >>