Monthly Archives: September 2018

The Graduate Interdisciplinary Earth Science Symposia: a year in review, and looking to the future

Below is an article I wrote for our department newsletter about the GIESS symposia. I’m publishing it here because it didn’t make the cut for the newsletter in this cycle, and it’s kind of a time-sensitive article because it justifies moving forward with GIESS in the current format.

There is a short description of what GIESS is in the third paragraph.

———————————————————————————————————————————–

The Graduate Interdisciplinary Earth Science Symposia:
a year in review, and looking to the future

Andrew J. Moodie

Scientists are typically effective written communicators, since professional success in academia is so closely linked with funded grant proposals and published manuscripts. However, oral communication skills are frequently considered subordinate and not consciously developed and practiced by early-career scientists.

In the summer of 2017, a group of EEPS graduate students addressed this training gap and the shortcomings of a weekly departmental seminar series by launching the Graduate Interdisciplinary Earth Science Symposia, or GIESS.

GIESS (pronounced “geese”) provides a forum for oral communication skill development while simultaneously encouraging department-wide participation in advancing the presenter’s science. The GIESS committee’s plan was to decrease the burden from weekly to monthly, limiting the seminar to two speakers, and intentionally adding prestige to presenting, thereby increasing the quality of the talks. Furthermore, the plan introduced “pop-up talks”, brief presentations allowing no more than two slides for no more than two minutes, as a lower-stakes opportunity to practice oral communication. Finally, lunch would be provided for participants to bolster attendance, and year-end awards would be given to select speakers and participants. Finally, department members were invited to provide feedback through an online survey at the end of the Spring semester to assess the inaugural year of the GIESS. The survey garnered 22 responses, predominantly from graduate students (19 of 22, 86%).

a) respondents overwhelmingly agreed that the GIESS was an improvement over the old seminar format. b) a paired t-test determines at above the 99% confidence level that the perceived quality of talks was improved by the GIESS format (p = 0.0001), based on respondents’ declared average quality of talks in the old seminar format and in the GIESS format on an integer scale from 1–10. c) respondents agreed that the pop-ups were a good addition to the GIESS. d) unsurprisingly, respondents were very happy with lunch. It should be noted that the survey designer (that’s me) has no training in survey design and the sample size is small, so the results presented herein should not be expected to withstand rigorous methodological or statistical scrutiny.

 

Overwhelmingly, participants agreed that the GIESS was an improvement over the old Friday Seminar format (Figure 1a, 68%), suggesting that the committee’s primary objective to improve the program was achieved. More importantly, the committee met the goal to improve the quality of the talks. A paired t-test determines at above the 99% confidence level that the perceived quality of talks was improved by the GIESS format (Figure 1b, p = 0.0001), based on respondents’ declared average quality of talks in the old seminar format and in the GIESS format on an integer scale from 1–10. The mean score given to talk quality improved from 4.9 to 7.6 between the old format and the GIESS format, respectively. The committee intends to have the presentation format in the GIESS remain largely the same in the 2018–19 year, possibly tweaking the duration of each talk based on respondent feedback.

Pop-up talks were a new addition, so a close look at participant opinions is warranted. Pop-up talks were scheduled between the two main presentations of the meeting, with typically three to five people participating. The addition of a short presentation style was decidedly favored (Figure 1c), with more than 80% of respondents agreeing that pop-ups were a positive addition. However, three respondents (15%) felt strongly that pop-up talks were a bad idea, though none clarified their position in an open-ended response. Therefore, the committee will keep pop-ups as a permanent feature of the GIESS. The committee also surveyed respondents about allotting time for audience questions of pop-up presenters. Seven respondents stated there should be no questions, seven were neutral on the issue, and seven think it would be a good idea; in the 2018–19 year, the committee will explore allowing questions directed at pop-up presenters.

Unsurprisingly, when more than 80% of survey respondents are hand-to-mouth graduate students, offering a free lunch to participants was favored; 20 out of 21 survey respondents agreed that providing lunch was a positive (Figure 1d). Attendance at the GIESS was vastly improved from the old seminar format, however, the number of students and faculty in attendance faltered late into the year. The committee tentatively attributes the increased attendance to lunch, though hopefully participants also came for the programming. In specific lunch-related feedback, one student asked for a “dessert table” next year, and more than one respondent added that they would like to see “platters of Chick-fil-A spicy chicken sandwiches with tangy Polynesian sauce.” Noted.

Finally, the GIESS committee would like to recognize the award winners who gave outstanding presentations, creative pop-ups, and engaged throughout the symposia. Best Talk awards for the year go to Brandee Carlson and David Blank, whose research presentations are respectively titled “Tie channels on deltas: A case study from the Huanghe (Yellow River) delta, China” and “Discrete element method as a tool for simulating megathrust earthquakes: insights into stress transfer”. Chenliang Wu and Cailey Condit received the Best Pop-up awards for presentations that were both fun and informative. The Best Participant awards were given to two first-year students Michael Lara and Patrick Phelps, because of their active engagement as participants in all aspects of the GIESS.

I would like to directly thank the committee members who helped make this inaugural year of the GIESS an outstanding success: Laura B. Carter, James Eguchi, Sahand Hajimirza, Harsh Vora, and Daniel Woodworth. Let’s make it even better next year.

 

Predicting equilibrium channel geometry with a neural network

In an attempt to learn more about ML I decided to just jump in and try a project. Predicting channel geometry with a simple neural network.

[source code]

All in all, I’m fairly sure I didn’t learn anything about equilibrium channel geometries, but I had some fun and learned an awful lot about machine learning and neural networks. A lot of the concepts I have been reading about for weeks finally clicked when I actually started to implement the simple network.

I decided to use the data from Li et al., 2015 [paper link here], which contains data for 231 river geometries.

The dataset has variable bankfull discharge, width, depth, channel slope and bed material D50 grain size.

We want to be able to predict the width, depth, and slope from the discharge and grain size alone. This is typically a problem, because we are trying to map two input features into three output features. In this case though, the model works because the output H and B are highly correlated.

The network is a simple ANN, with one hidden layer with 3 nodes. Trained with stochastic gradient descent. Training curve below.

Matlab speed comparison of switch-case and if-then statements and hard-code

I frequently use the extremely methodical approach in scientific programming of “just trying things”. This means that I create a lot of different ways to try to do something to find the best way to do that thing. Sometimes this means getting a realistic result, a stable result, or a fast computation.

All functional programming languages offer if-then statements, whereby sections of code are evaluated on True evaluations of a statement. An if-then statement can be extended with an elseif to provide further evaluations to compare with for True. Using this framework, different “cases” or “methods” of doing a single thing can be quickly tested by switching between them in a script. The purpose of all is to make the code more readable and maintainable, by not having to comment/uncomment blocks of code to make a change in the calculation method.

For example, two methods x = 'test1' | 'test2' could be executed by a function, depending on the definition of x:

if strcmp(x, 'test1')
y = 1;
elseif strcmp(x, 'test2')
y = 2;
end

A similar functionality can be obtained with a switch-case statement:

switch x
case 'test1'
y = 1;
case 'test2'
y = 2;
end

But which is faster?? I suppose I always knew switch-case statements were slower than if-then statements, but I wanted to know how much of a difference it could make. I also often have >4 of these case statements as optional methods, and I wanted to know if the number of cases (or really how “deep” down the list they were) made a difference in speed.

I designed a simple experiment in Matlab to test this out. I looped through a simple set of statements 1 million times, and timed each scenario. You can find the source code here.

It turns out that switch-cases are about ~30% slower than if-then statements. Both are more than an order of magnitude slower than.

Most importantly though, the time increases linearly with each “layer” to the if-then or case-switch statement. To me, this stressed the importance of 1) having few cases that aren’t simply hard-coded in, or 2) at least sort the cases in order of likelihood of being used during program execution.