science

discussion

New Blog - First results at AAS!

Started by jules
Default_user
almost 7 years ago

Team member Cliff Johnson has presented the first Andromeda Project results at the American Astronomy Society meeting. Download the poster!

Default_user
almost 7 years ago

I get the impression that this is 'high speed science', from ('big') data to results in record time ... and it is we Citizen Scientists who made it happen.

AWESOME! :-)

Bring on Round #2!!

Default_user
almost 7 years ago

Precisely! The team are overjoyed at the speed and enthusiasm of the AP citizen scientists! Hubble is currently working hard on taking the next batch of images. So there is more to come. :)

Default_user
almost 7 years ago

And the poster has your name on it jules, as an author! Well done!! :-D

Default_user
almost 7 years ago

Just behind Cliff's left shoulder (his; it's on the right from our perspective) is part of the poster entitled "Initial User Drawings". In the poster you can read (just) in the top left corner "B15-F03_3 / Views: 81"; the image to the right - "Final Merged Catalog" - lacks this label.

What, you may be wondering, is this AP image?

Well, I don't know how to convert from JPEG files names to "AAP0000nnn" IDs (but I'm working on it), but here is that field:

And here it is in B&W:

Based on Cliff's comments - in the poster - about the detection threshold for young, low mass clusters, quite a few of the "few views" cluster-candidates may be just this kind of cluster! And the reason why so few AP zooites marked these may be due more to the definitions in the Guide than to an inability of zooites to spot clusters (e.g. "a true star cluster will be tightly grouped with at least three or four individually visible stars", and "Because of this, one reliable signature of a star cluster is an enhancement in the background light, giving them a “fuzzy” appearance. This fuzz may be less apparent in the more extended diffuse clusters, where more of the faint stars can be detected individually.").

Maybe the guidelines could be revised, for Round #2?

Default_user
almost 7 years ago

Maybe - though I'm not sure I would have ID'd SC5182 as a cluster at all. B&W version needed there but even so...

And thanks - very happy to be included as an author!

Default_user
almost 7 years ago
JeanTate in response to jules (moderator)

jules:

Maybe - though I'm not sure I would have ID'd SC5182 as a cluster at all. B&W version needed there but even so...

And this goes pretty much to the heart of the point I am trying to make: by construction - and therefore by ('true') definition - SC5182 is a cluster; by the Guide's definition, it pretty obviously isn't.

Knowing - now - that it really, truly, cross-my-heart-and-hope-to-die (or Cliff's anyway ;-)) is a cluster, we AP zooites have to adjust our perceptions of what a 'cluster' is, in terms of what we train our eyes+brains to search for.

At first glance, it might seem that this re-training would result in a large number of false positives (and even after some trials that may still be true); however, the key thing about us humans is that we can adapt, in ways robots and computers can't (hopefully).

It may turn out that these low-mass, young clusters cannot be reliably found, even by highly trained humans; if so, maybe a different kind of image might work? For example, taking a cue from PH, maybe we could be served up CMDs (color-magnitude diagrams) of 'low views' cluster candidates? We'd then look for <s>transits</s> clusters in the patterns in the charts (or do you say 'graphs'?). I'm brainstorming of course, but I think you get the idea ...

And thanks - very happy to be included as an author!

You're very welcome!

Default_user
almost 7 years ago
jules (moderator) in response to JeanTate

Agreed - that's what I meant (but didn't make clear!) Even with extra training to point out that some clusters can look like SC5182 I would have had difficulty committing to it. The likelihood of false positives as you say is too great unless we have additional tools. I am all for additional tools!

Default_user
over 6 years ago

Fantastic project, very well led and with terrific support! Congratulations on the work done so far and thanks very much for the feedback. :)

In the case of SC5182 I would probably not have marked the cluster as such from the colour image; the definition we were given to work with excluded it imho. I think that it may be helpful for the next phase to include a working annotation tool or at least an "other interesting object" button. If we can identify precisely which area we are trying to highlight we can then open a discussion thread to form a collective opinion on potential cluster identification? Even if only a small number of users got involved it may effectively expand detection thresholds without compromising the wisdom of crowds philosophy??

Default_user
over 6 years ago
jules (moderator) in response to mjtbarrett

I'd agree with that. It shouldn't affect the classifications as we would likely have to discuss images after classifying them and thereby be better equipped to tackle the next tricky one. Anything to increase accuracy has to be good.

Please Log In to make comments.