Graduate Exam Abstract

John Hall

Ph.D. Preliminary
July 6, 2020, 11:00 am - 1:00 pm
Long-Term Learning for Adaptive Underwater UXO Classification

Abstract: The problem of underwater object classification in sonar imagery is rather complicated due to the numerous factors which inhibit repeatable and reliable Automatic Target Recognition (ATR). These factors include: variations in operating and environmental conditions, presence of spatially varying clutter, as well as burial depth and variations in target shapes, compositions and orientation. In the context of lifelong learning, these constantly changing environmental factors can pose many obstacles to efficient and confusion free learning of targets and surrounding environments. A lifelong learning machine is a machine that learns continuously, accumulates the knowledge learned in previous tasks, and uses it to help future learning. This continuous learning must occur without negatively impacting the learning in previous environments. That is, the learner accumulates more pertinent knowledge in relation to its tasks while offering stability of prior knowledge base. To further complicate matters, access to extensive training datasets is nearly impossible due to the nature of the problem and many datasets that exist are completely unlabeled or only partially labeled.
In my continuing work on classification for shallow water sonar systems, I propose to address the following fundamental questions regarding lifelong learning for classification of UXOs in sonar data:
1. What are the most viable mechanisms to allow an unmanned underwater vehicle to accumulate and incorporate novel labeled or un-labeled data into its target identification system without sacrificing performance in old environments?
2. What are the most viable mechanisms for allowing an underwater UXO minehunter to extract class labels despite varying environmental conditions?
3. What are the advantages, shortcomings, and major differences, of compressed-sensing based approaches to target identification, such as the modified matched subspace classifier (MSC) with incremental dictionaries, versus multi-task learning approaches?
4. How do other state-of-the-art inference systems address the lifelong learning problem and how they fair against the proposed solutions specifically for the underwater target classification problem?
The topic of this preliminary presentation will focus on developments made to a sparse form of the MSC framework which allows the model to incorporate novel important samples when they are discovered from a new environment. Specifically, a method is introduced which allows incremental updating of a kernelized embedding that depends of far fewer samples than a full kernel solution while retaining a useful and consistent geometric interpretation. The preliminary results and some of the key features of the proposed solutions will also be presented.

Adviser: Dr. M. R. Azimi-Sadjadi
Co-Adviser: N/A
Non-ECE Member: Dr. Michael Kirby, Mathematics
Member 3: Dr. Ali Pezeshki, ECE/Math
Addional Members: Dr. J. Rockey Luo, ECE

M. R. Azimi-Sadjadi, C. Robbiano, Y. Zhao, J. Hall. "Incremental Dictionary Learning For Adaptive Classification And Reconstruction Of Facial Imagery." 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP). IEEE, 2019.

N. Larson, J. Hall, and M. R. Azimi-Sadjadi. "Analyzing Transfer Learning Methods For UXO Classification In Varying Shallow Water Environments." 2019 IEEE 29th International Workshop on Machine Learning for Signal Processing (MLSP). IEEE, 2019.

J. Hall, M.R. Azimi-Sadjadi, and S. Kargl, “Underwater Unexploded Ordnance (UXO) Classification Usinga Matched Subspace Classifier With Adaptive Dictionaries ”IEEE Journal of Oceanic Engineering, June 2018.

Y. Jiao, J. Hall, and Y. Morton, “Automatic GPS Ionospheric Amplitude and Phase Scintillation Detectors Using a Machine Learning Algorithm”Inside GNSS Magazine, May 2017.

Y. Jiao, J. Hall, and Y. Morton, “Performance Evaluation of an Automatic GPS Ionospheric Phase Scin-tillation Detector Using a Machine Learning Algorithm”Journal of The Institute of Navigation, March2017.

Y. Jiao, J. Hall, and Y. Morton, “Automatic Equatorial GPS Amplitude Scintillation Detection using Ma-chine Learning”IEEE Transactions on Aerospace and Electronic Systems, February 2017.

J. Hall, M.R. Azimi-Sadjadi, and S. Kargl, “Underwater UXO Classification using Matched Subspace Clas-sifier with Synthetic Sparse Dictionaries,”IEEE/MTS Oceans ’16 Conference September 2016.

Y. Jiao, J. Hall, and Y. Morton, “Performance Evaluations of an Equatorial GPS Amplitude Scintillation Detector Using a Machine Learning Algorithm”Proceedings of ION GNSS+ Conference, September 2016.

P. Jamieson, L. Grace, J. Hall, and A. Wibowo “Metaheuristic Entry Points for Harnessing Human Compu-tation in Mainstream Games”HCI International, July 2013.

P. Jamieson, L. Grace, and J. Hall, Research Directions for Pushing Harnessing Human Computation toMainstream Video Games,Meaningful Play Conference, October 2012.

Program of Study:
MATH 676
STAT 620
STAT 525
ECE 511
ECE 520
ECE 795
ECE 795
GSTR 600