Events

Learning Structured Signals using Active, Robust, and Ensemble Methods
Friday, May 18, 2018 - 12:30pm

What: Computational & Data Science Seminar

 

When: 12:30 pm May 18th, 2018

 

Where: Cramer Hall 418

 
Title: Learning Structured Signals using Active, Robust, and Ensemble Methods
 
Abstract: Modern problems in signal processing and machine learning involve the analysis of data that is either high-volume or high-dimensional. In one example, scientists studying the environment must choose their set of measurements from an infinite set of possible sample locations. In another, performing inference on high-resolution images involves operating on vectors whose dimensionality is on the order of tens of thousands. To combat the challenges presented by these and other applications, researchers rely on two key features intrinsic to many large datasets. First, large volumes of data can often be accurately represented by a few key points, allowing for efficient processing, summary, and collection of data. Second, high-dimensional data often has low-dimensional intrinsic structure that can be leveraged for processing and storage. In this talk, I will show how we can leverage these facts to develop and analyze algorithms capable of handling the challenges presented by modern data.
 
First, I will discuss the problem of determining regions of low oxygen concentration (hypoxic regions) in lakes via an autonomous surface vessel. Hypoxic regions are non-stationary, and hence the sampling procedure must be performed in as little time as possible. I will present an active learning algorithm called quantile search that accounts for sampling cost in terms of both number of samples taken and distance traveled. Simulations and experimental results show the algorithm significantly outperforms state-of-the-art methods in terms of total sampling time. Second, I will cover the topic of subspace clustering, i.e., determining a union of low-rank subspaces that best fits a high-dimensional dataset. This can be viewed as a generalization of principal component analysis and is an increasingly popular model for clustering high-dimensional data, with applications in facial recognition and object tracking. I will discuss algorithms for solving this problem in both the unsupervised and semi-supervised settings.
 
Bio: John Lipor is an Assistant Professor of Electrical Engineering at Portland State University. He earned his BSEE from the University of Wisconsin-Madison, MSEE from the King Abdullah University of Science & Technology (KAUST), and PhD from the University of Michigan. John is an awardee of the NSF Graduate Research Fellowship. His research interests include (inter)active learning and subspace methods, with an emphasis on environmental sensing and monitoring.
 
Following the talk, at 1:30, in the same room:
Will Garrick will lead a hands-on workshop on using Matlab on PSU's HPC cluster. Interested parties are encouraged to request access to the HPC cluster prior to the session, and to bring their laptops to the event so they can set up everything necessary to begin using cluster resources.
For more info about this workshop, email: oit-rc-group@pdx.edu.
 
For more info about the overall seminar series:
 
To add this seminar to your calendar:  calendar link
 
Please forward this announcement to students and colleagues whom you think might be interested in this seminar series, especially anyone who uses Matlab and might want to attend the 1:30 workshop with Will Garrick. Thanks!
 
Wayne Wakeland and Bruno Jedynak