Hello everyone, my name is Rachel Meyer, and I am a student at Olivet Nazarene University. I have a professor who was an ACEAP Ambassador and who suggested that I should reach out to people who are involved with the LSST, for information within my research field.
I am an engineering student in the honors program, and I am currently working on a research project that has to do with the use of machine learning in astronomy, specifically asteroid detection. This is a two-year project and I started working on it this semester, and I am super excited to continue my research, and I am looking for any information on this topic that anyone would be willing to share. I am planning to use deep learning in my research to detect asteroids, and hopefully in the future I will be able to create my own system to help with the detection of specific asteroids. I was wondering if there were any resources that would be helpful for my research or with the use of machine learning at Vera Rubin, as I understand machine learning will play a large role in analyzing the data that the telescope generates.
If you want to know more about me, some things I do at school involve working at the circulation desk in our library, as well as an operator at our on-campus planetarium. I am also heavily involved in our chapter of SWE and this year I am on the council as the Vice President. I am also on the council for the Honors Student Council as a Co-Chair, which is the program that I am doing my research through. One of my favorite things to do is be a teaching assistant for the astronomy labs at my school, and I really enjoy showing my classmates how to use telescopes and calculate properties about space!
Hi Rachel, thanks for posting here. Asteroids aren’t my field, but your request reminded me of this deep learning project to detect asteroids with the Zwicky Transient Facility (ZTF), which is a wide-area sky survey similar to the future Legacy Survey of Space and Time (LSST) with the Rubin Observatory.
Happy to see you here. Most of our work so far in terms of detecting asteroids with Rubin in the LSST pipelines has been focused on developing more deterministic approaches to measurement and linking, but there are certainly places where machine learning is helpful. One of those is in identifying streaks, another would be in separating real and artifact sources in difference images, and others would be in identifying activity in images of asteroids or in just identifying outliers in the catalogs of objects that are found or possibly in identifying outliers in astrometric or photometric measurements associated with a particular asteroid.
Because we’ve been focused on deterministic approaches first, I’m not sure what we have that would be most useful to you. One of the exercises we have gone through was creating simulated catalogs of asteroid detections, so we could test our linking methods. Did you have something in particular you were interested in or looking for?
Some of the people in the project who might be interesting to talk to include @mjuric or @eggl … some of the people from the science collaboration might include @hhsieh. I’m tagging them here in case they have anything they’d like to add.
First, to echo Melissa and Lynne, it’s great to see you here. Regarding your questions, as Lynne discussed, baseline asteroid detection algorithms in LSST data are already pretty far along in development, so work that you could help with would probably be in more specialized cases (which Lynne gave several good examples of). As one of my main personal scientific interests is in active objects (i.e., things that look like comets), I will point out that activity detection is an area in which machine learning could probably help a lot since it’s a pretty complex problem because it turns out that a lot of image artifacts look like comets and vice versa, and so machine learning could help to sort those out. The problem of course is getting a good training set, which is especially difficult for LSST considering that it’s not operational yet and so no real LSST data is currently available. One workaround for that is that you could use data from existing surveys to start working on your training set-up so that when LSST is up and running, you’re ready to go. @mkelley might be able to help you with accessing ZTF data for that purpose. Alternatively, if you are more interested in trying to apply machine learning to linking different asteroid detections together, the simulated asteroid detection catalogs that Lynne mentioned would probably be the way to go. But yes, as Lynne said, if you can give us a bit more information about what you want to do, we can get a better of idea of which direction to point you.
I have read a little bit about the Zwicky Transient Facility in other research papers but I have yet to read this one, so thank you for sending it to me, it looks like it will be very useful and informative for my research!
@ljones@hhsieh Hi Lynne and Henry, thanks for responding! I am not entirely sure where my research is headed at this point and I am mostly exploring this field to see what holes there are that I can fill and do more research on. I know that there is already a lot done in this field, so I am looking for specific problems to work on, which you both have provided some very good examples that I am going to have to look into! As of right now I am looking at the preprocessing stage for asteroid detection and seeing what I can do there, with detecting outliers and other unimportant variables.
One workaround for that is that you could use data from existing surveys to start working on your training set-up so that when LSST is up and running, you’re ready to go.