IEEE OC Section & Computer Society
Meeting Date: June 28, 2017
Time: 6:00 PM Networking & Food; 6:30 PM Presentation
Speaker: Dr Joshua Harguess of SPAWAR Systems Center Pacific
Location: Qualcomm San Diego
Event Details: IEEE vTools
Summary: Deep learning has continued to gain momentum in applications across many critical areas of research in computer vision and machine learning. In particular, deep learning networks have had much success in image classification, especially when training data is abundantly available. However, researchers have shown vulnerabilities to these deep learning networks that are not well understood. For instance, when estimating optical flow from real-world video data, such as that taken of full-motion-video (FMV) from unmanned vehicles, surveillance systems, and other sources, the results from deep learning and other computer vision approaches are often undesirable. Additionally, several researchers have exposed potential vulnerabilities of deep learning networks to carefully crafted adversarial imagery. In this talk, we’ll examine the effects of real-world video data on motion estimation and a potential path forward to detecting adversarial imagery using image quality metrics.
Bio: Dr Joshua Harguess
*PhD Electrical and Computer Engineering, 2011, University of Texas Austin,
Portfolio in Applied Statistical Modeling; MS Computational & Applied Mathematics, 2007, University of Texas Austin,
*Research Scientist, SPAWAR Systems Center Pacific, San Diego,
*Successfully managed and executed on the continuing Independent Applied Research project Video notation, Search and Tagging (VAST).
*Active member of the Motion Imagery Standards Board (MISB) and image and video quality working group.
* In the Automated Imagery Analysis group, performed research on ship classification using state-of-the-art mthods and the effects of pre-processing (SPIE 2012), anomaly detection for ship detection and tracking from full-motion video from UAV. Most recently, developed new algorithm using sparse representation for anomaly detection with applications in many domains.
*Performed research into using evolutionary techniques to train a deep learning network to learn features for classification of maritime imagery (ICLR 2014 & GECCO 2014).
* In the Unmanned Systems group, performed research on automatic calibration between stereo and infrared (IR) cameras for robot vision and navigation (SPIE2014), motion estimation and optical/range flow for robot ego-motion estimation (PLANS2014), and studying the effects of optical flow on degraded imagery.
*Co-founder of both the Computer Vision and the Machine Learning Series group at SSC Pacific which consists of several researchers across the center that meet biweekly to discuss computer vision and machine learning research