February PDXML Meetup

Same place, still looking for speakers. If you have anything you would like to present, let me know!

If you need parking, there's a parking deck below Safeway.

Agenda:

6:00 p.m.: Food, beverage, and networking

6:40 p.m.: Welcome message by Karl Fezer

6:45 p.m: Speaker 1: "Strategies for handling missing data in modeling and prediction" by Robert Dodier
7:30 p.m: Speaker 2: "Data Science View of Subduction Zone Imaging" Frederick Pearce

8:15 p.m.- 8:30: Project Ideas. Pitch your Project Ideas to this meet-up group

8:30: End

Speaker 1 Details:
Title: "Strategies for handling missing data in modeling and prediction"

Abstract: Machine learning models are typically phrased in terms of
input and output data, and, often enough, some data have missing
(i.e., unknown) values. I'll talk about some common missing data
scenarios, and present a general Bayesian framework for working with
missing data. I'll show how solutions for specific scenarios can be
derived from the general framework, and I'll note how conventional
heuristics for missing data can be understood from a Bayesian point of
view. Finally I'll present a few worked examples to illustrate
specific problems and their solutions. This presentation will be light
on the math, with an emphasis on the many interesting concepts in play
here. Bio: Robert Dodier has a background in machine learning and data
science. He holds a PhD in Civil Engineering from the University of
Colorado where his dissertation was on Bayesian network models applied
to engineering problems. Since then, Robert has worked on a diverse
range of problems, including modeling customer behavior in the
telecommunications industry, development of a system of autonomous
software agents, and, most recently, analyzing and modeling electrical
loads for an automated demand response system.

Speaker 2 Details:
Abstract: "Data Science View of Subduction Zone Imaging"
Fred will go over data science view of his MIT Ph.D. thesis work, including a description of the data, the preprocessing/imaging methods, and the interpretation of the resulting seismic images in terms of subduction zone structure. The data preprocessing employs a novel unsupervised machine learning pipeline that computes high accuracy (and precision) receiver functions by iteratively applying multichannel cross-correlation, principal component analysis, and frequency-domain deconvolution.
If you want to view the slides, they're available here: https://www.slideshare.net/FrederickPearce/data-science-view-of-seismic-imaging

Bio: Fred Pearce is a data scientist with an extensive technical computing and research background honed during his many years at the Massachusetts Institute of Technology (MIT) and Los Alamos National Laboratory. He holds a Ph.D. and M.S. from MIT in Geophysics and Geosystems, respectively. Fred has worked on a wide range of problems, including the integration of seismic images and geodynamic models to better understand subduction zone processes, how hysteretic soil behavior impacts ground motion during large earthquakes, developing novel features for flagging anomalous events in security
data, and many others.