I recently came across this site http://www.planethunters.org. It shows brightness observed for a star vs time. It asks questions like if star is exhibiting variable or fixed pattern. Later it asks if there are any transits which can indicate presence of planet.

While I find this interesting, it made me wonder why can't a computer program detect these patterns. Is this really hard? Given the fact this field has been subject of research for past so much time. These days we have programs can detect face, read text on live phone camera and translate it. So what's stopping this?

P.S. Also how can I ask same question across multiple stackexchange site? Maybe this is more SO / Physics question.

  • $\begingroup$ For now (in physics.SE), it is off-topic. It maybe good for Programming.SE $\endgroup$ – Waffle's Crazy Peanut Dec 29 '12 at 4:33
  • $\begingroup$ I've voted to reopen, because it appears the reason for it being closed was simply that it happens to mention computer programs. However, this alone does not make it off topic according to the FAQ and I believe it should not have been closed. I would suggest that you (Ankush) re-write the question to be more along the lines of "what is it about the signal that makes the patterns difficult to detect?" If you avoid mentioning computer programs it will quite likely be reopened. $\endgroup$ – Nathaniel Dec 29 '12 at 9:44
  • $\begingroup$ @Nathaniel No, this question was not closed because it happened to mention computer programs. We don't close questions because they happen to mention one thing or another, we close them because they are not about physics. This one is about the limitations of algorithmic analysis. Rewriting it in the way you suggest wouldn't make it on topic here. $\endgroup$ – David Z Dec 29 '12 at 9:51
  • $\begingroup$ @DavidZaslavsky Eh? The reason you can't easily write a computer program to detect these signals is because there is a very low signal to noise ratio, and the reasons for that are entirely to do with physics, e.g. the distances involved, the tiny size of a planet compared with a star, etc. It seems quite clear to me that this is what's being asked about, and it's quite clearly on topic. $\endgroup$ – Nathaniel Dec 29 '12 at 10:00
  • 1
    $\begingroup$ The question is about processing a signal from an astronomical instrument. Signal processing isn't within our scope. Once the question gets to the point of interpreting what has been processed out of that signal, then it'd be on topic here, but at its current stage, or as what you're suggesting it be rephrased to, it would go on another site, possibly Signal Processing or Stack Overflow depending on what sort of processing is being done. $\endgroup$ – David Z Dec 29 '12 at 11:01

The computer program can be very simple and they have existed in one form or another for many years. At least for the purpose of defining variable star pulsation timings.

The problem lies in both in signal to noise ratios and total timings.

The variation in light can be lower than milli-magnitudes for planetary passes and as such this is very tricky to measure at ground level. The Kepler sattelite is currently monitoring thousands of stars above the atmosphere and so has a higher detection potential.

And to measure the timings accurately you want as many passes as possible. But if the orbit is of the order of 100's of days this can mean watching a star for years! Therefore the data available for analysis is very patchy and of low 'quality'. But we're getting better and more telesccopes are being aimed at potential targets so keep watching!


It's because present-day computers are syntactic devices, not semantic devices. Humans use both levels (and possibly others) to figure out our way in this world, whereas computers only have access to the first level.

More simply put, computers lack the big-picture perspective, and the hugely interconnected network of representations that our brains are so good at. Sometimes you need that to figure out whether this small bump in the graph is a mere random event, or is it meaningful (semantic level) data.

Artificial intelligence will really take off when we figure out how to do semantic computing.


That is a very good question.

I think the answer is that it is a quest for the unexpected. Programs have been developed to detect the transits, but perhaps there are new kinds of discoveries waiting to be made by having thousands of people looking at the data. From Humans vs. Machines on the Planet Hunder's site:

... developing computer algorithms to analyze light curve data ... we expect computer programs to robustly identify things that they are trained to find ... will be a number of surprises in the data that the computer algorithms will miss.

The human brain is particularly good at discerning patterns or aberrations and experiments ... collective wisdom ... can be better than an expert ... taps into the power of human pattern recognition ... new classification schemes for different families of light curves, identify oddities, and verify transit signals.

A particular example is Two exoplanets discovered by “citizen scientists”:

Remarkably, both stars in this case were flagged as potential planet-bearers by the software, but also removed from the list to make followup observations by the same code!

In one of those two cases, the human detection was used to find a false negative.


Not the answer you're looking for? Browse other questions tagged or ask your own question.