Ian Fasel | University of California, San Diego (original) (raw)

Papers by Ian Fasel

In order for learning agents to be useful to non-technical users, it is important to be able to t... more In order for learning agents to be useful to non-technical users, it is important to be able to teach agents how to per- form new tasks using simple communication methods. We begin this paper by describing a framework we recently de- veloped called Training an Agent Manually via Evaluative Reinforcement (TAMER), which allows a human to train a learning agent

Abstract—There,is strong experimental,evidence,that new- born infants orient towards,human,faces ... more Abstract—There,is strong experimental,evidence,that new- born 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

Proceedings of the 2009 international conference on Multimodal interfaces - ICMI-MLMI '09, 2009

Lecture Notes in Computer Science, 2011

We describe a Bayesian network implementation of a theory of concepts that is motivated by observ... more We describe a Bayesian network implementation of a theory of concepts that is motivated by observations from the philosophical debate between Lexical Concept Empiricism and Lexical Concept Nativism. According to our theory, Baptizing Meanings for Concepts (BMC), concepts are acquired by hypothesizing latent kinds/categories to explain observed co-occurrences of sets of properties in a group of objects. The hypothesized kind/category

We present results on a user independent fully automatic system for real time recognition of basi... more We present results on a user independent fully automatic system for real time recognition of basic emotional expressions from video. The system automatically detects frontal faces in the video stream and codes them with respect to 7 dimensions: neutral, anger, disgust, fear, joy, sadness, surprise. The face finder is based on (18) with a more complex fea- ture space and multiframe exclusion rules. The expression recognizer receives image patches located by the face detector. A Gabor represen- tation (2) of the patch is formed and processed by bank of 63 SVMs (3). The final coding into 7 expression categories is performed via multino- mial ridge logistic regression, a natural generalization of SVMs to the multinomial case. Strategies for performing multiclass decisions using SVM's are compared. The effectiveness of Gabor magnitude filters is examined. Different methods for combining information from the upper and lower regions of the face are also discussed. Results on the Cohn- ...

There is currently a gap in automatic facial expression recognition between the levels of perform... more There is currently a gap in automatic facial expression recognition between the levels of performance reported in the literature and the actual performance in real life conditions. A troublesome aspect of this gap is that the algorithms that perform well on the standard datasets and in laboratory demonstrations could be leading research in the wrong direction. To investigate this issue, we document the process of developing a smile detector for real world applications. We thoroughly explore the required characteristics of the training dataset, image registration, image representation, and machine learning algorithms. Techniques from the psychophysics literature are presented for detailed diagnosis and refinement of the obtained smile detector. Results indicate that current machine learning methods are appropriate for developing real-world expression recognition systems provided that: (1) The right combination of classifier and feature sets is selected, and (2) a sufficiently large (...

7th International Conference on Automatic Face and Gesture Recognition (FGR06), 2006

2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), 2005

2004 IEEE International Conference on Systems, Man and Cybernetics (IEEE Cat. No.04CH37583), 2004

2007 7th IEEE-RAS International Conference on Humanoid Robots, 2007

2003 Conference on Computer Vision and Pattern Recognition Workshop, 2003

Robotics and Autonomous Systems, 2009

Most humanoid soccer robot teams design the basic movements of their robots, like walking and kic... more Most humanoid soccer robot teams design the basic movements of their robots, like walking and kicking, off-line and manually. Once these motions are considered satisfactory, they are stored in the robot’s memory and played according to a high level behavioral strategy. Much time is spent in the development of the movements, and despite the significant progress made in humanoid soccer

2010 IEEE 9th International Conference on Development and Learning, 2010

ABSTRACT Training deep belief networks (DBNs) is normally done with large data sets. In this work... more ABSTRACT Training deep belief networks (DBNs) is normally done with large data sets. In this work, the goal is to predict traces of the surface of the tongue in ultrasound images of the mouth during speech. Performance on this task can be dramatically enhanced by pre-training a DBN jointly on human-supplied traces and ultrasound images, then training a modified version of the network to predict traces from ultrasound only. However, hand-tracing the entire dataset of ultrasound images is extremely labor intensive. Moreover, the dataset is highly imbalanced since many images are extremely similar. This work presents a bootstrapping method which takes advantage of this imbalance, iteratively selecting a small subset of images to be hand-traced, then (re)training the DBN, making use of an entropy-based diversity measure for the initial selection. With this approach, a three-fold reduction in human time required to trace an entire dataset with human-level accuracy was achieved.