NASA SOHO Comet Search with Artificial Intelligence Open-Science Challenge

Something a little different here...

The "NASA SOHO Comet Search with Artificial Intelligence Open-Science Challenge" has officially ended, and ended with great success, including two new SOHO comet discoveries

Earlier in 2022, you may recall that the Sungrazer Project, sponsored by NASA, partnered with LISH and TopCoder to present a novel 'crowd-sourced' Artificial Intelligence / Machine Learning (AI/ML) data challenge with a goal of developing algorithms that could automatically detect and track comets in LASCO C2 images. For the challenge, we provided over 3,000 data sets related to SOHO/LASCO C2 comet observations, with those data sets comprising the raw FITS science data and a plain text file with x,y positions of comets. Registrants for the contest were then tasked with developing AI/ML algorithms that could learn to detect and track comets in the C2 data. The contest ran for three weeks and received 212 submissions of algorithms, with the seven highest-scoring algorithms being awarded prizes. The codes were subsequently handed to NASA as open-source software. 

As Sungrazer PI, I was of course involved in developing the requirements of the contest and providing guidance on how the submissions should be scored. I was not actually involved in developing the algorithm scoring codes, or in ranking the winning algorithms - that was all down to a great team of folks at TopCoder. But following the conclusion of the contest, I was able to take a look at the output of the algorithms to see what kinds of features they were detecting. Of course, they were very effective at detecting even relatively faint comets, but also returned lots of "noise", as you would expect. (SOHO comet hunting is not easy!) 

Most of the data provided for the contest contained known SOHO discoveries, but we did also provide a number of "empty" data sets containing no known comets. The purpose of that was to see if the algorithms were able to tell when no comets are present, and of course the contest registrants were not told which data sets were "empty" of comets. But it turns out that one of the "empty" data sets was actually not empty, and contained a previously unreported comet! This was a very pleasant surprise, but an even bigger surprise was that one of the data sets containing a known comet also contained an unknown comet! So two new comets!

I am still working on getting the names for the discoverers of these new comets, so I do not have SOHO comet numbers for them yet. But I can provide some details. The first comet was a Kreutz-group comet found in data from 2005-05-06. There were several comets in the data around this time, so I had to check carefully that it was not a known object, but I cannot find any confirmed or unconfirmed reports that correspond to it. (It was a little off-track, and certainly not bright). Positions for that are as follows, with the (0,0) position in the lower-left

  DATE     TIME     TEL     COL     ROW
2005-05-06 01:31:49  C2   310.70   14.94
2005-05-06 01:54:05  C2   322.44   29.62
2005-05-06 02:06:07  C2   329.22   38.20
2005-05-06 02:30:05  C2   340.96   54.24
2005-05-06 02:54:05  C2   352.70   70.27
2005-05-06 03:06:06  C2   359.70   81.34
2005-05-06 03:30:05  C2   371.45   96.92
2005-05-06 03:54:05  C2   383.87  115.21
2005-05-06 04:06:05  C2   390.19  125.37
2005-05-06 04:30:05  C2   402.61  140.76
2005-05-06 04:54:05  C2   415.94  156.57
2005-05-06 05:06:05  C2   422.71  165.37

The second comet was a non-group comet in images from 2016-11-04. Again, I looked into all the archives and can find no mention of it, though there was a nearby Meyer-group comet in the same image set (this was clearly not a Meyer comet). Positions for that are as follows, again with the (0,0) position in the lower-left

   DATE     TIME    TEL    COL     ROW
2016-11-04 09:36:05  C2   408.54 1013.64
2016-11-04 09:48:05  C2   398.31  999.86
2016-11-04 10:00:06  C2   387.44  985.65
2016-11-04 10:12:05  C2   376.58  970.82
2016-11-04 10:24:05  C2   366.98  957.25
2016-11-04 10:36:10  C2   356.32  942.21
2016-11-04 10:48:05  C2   345.88  927.80
2016-11-04 11:00:05  C2   335.23  912.55
2016-11-04 11:12:05  C2   325.83  897.31
2016-11-04 11:24:05  C2   315.81  881.85
2016-11-04 11:36:05  C2   306.07  866.57
2016-11-04 11:48:05  C2   296.67  851.74
2016-11-04 12:00:05  C2   287.48  835.66

Developing new, fast, and efficient algorithms for handling, processing, and analyzing heliophysics (e.g. LASCO) data is really important and really useful. The purpose here is absolutely NOT to replace citizen scientists! (I can assure you that, from looking at the output of the algorithms, they have a long way to go to compete with human eyes!) But there is tremendous value in pushing the boundaries of data analysis and image processing, and in seeking input from people who are not necessarily experts in heliophysics data but are experts in developing advanced algorithms. The ability to detect and track any kinds of features in heliophysics (or any astronomical) data is absolutely something of value. So we are very excited that not only do we have some really good algorithms to explore, but that a couple of new discoveries were made along the way.

Of course, this does raise the question of "who gets credit if an algorithm finds a comet"? This is a gray area, but thus far the Sungrazer Project has always given credit to the person(s) reporting the comets, without consideration for how they located the comets. Very early in the project I know for a fact that many of the comets were discovered by 'comet detection algorithms' that people were running, and I know that even now this still happens. And that's ok! To me it does not do anything to diminish the fact that this is a citizen science project, that allows anyone in the world to make a real and valuable contribution to astronomy and science. The huge diversity of people of different ages, backgrounds and expertise, and the great many approaches that are made to comet hunting, are what make the project so unique, and that will never change.

So congratulations to the developers of these algorithms for the comets they have discovered! I will post "official" confirmation of them soon.