Galaxy Zoo 2

Galaxy Zoo is the worlds largest astronomy collaboration with over a hundred thousand collaborators. I mentioned Galaxy Zoo some time ago, but since then they have doubled the number of papers published from their collaboration and launched Galaxy Zoo 2. I want to discuss the impetus, implementation, and some of the results of the project here. It begins with the desire of astronomers to classify galaxies. And why would you want to classify galaxies? Well in order to learn about the cosmos, that is to learn about the properties of merging galaxies in the local universe, to learn about galaxy star formation, or to learn about the intrinsic spin of galaxies in our universe you may need to classify galaxies. The Galaxy Zoo collaboration has implemented the citizen science or crowd sourcing model in order to classify the million or so images of galaxies taken by robotic telescopes (like the Sloan Digital Sky Survey, SDSS, which produces images just like those seen here). Many years ago astronomers only had to inspect astronomical images by eye from photographic plates. The digital revolution has brought computers to bear on the problem as astronomers have implemented machine learning techniques such as neural networks in order to identify spiral from elliptical galaxies, but approaches like nueral networks are still limited by their training set size and are prone to errors. So as robotic survey telescopes have allowed us to gather a fantastic amounts of data technology has not been so successful at making sense of that data.

There have been several successful models for citizen science at home such Seti at Home and Folding at Home, but the most powerful computational device most people have at home is their own mind. The wisdom of crowds had already come to bear on one astronomical project, Stardust at Home, so galaxy classification was a natural application for citizen science. The Galaxy Zoo team had a simple approach to galaxy classification they implemented in Galaxy Zoo 1. They offer the user a single image, like those seen above, of a galaxy and 6 buttons:

The buttons are as follows: 1) Elliptical galaxy, 2) Clockwise spiral, 3) Anti-clockwise spiral,  4) Spiral Galaxy Other e.g. Edge on, Unsure, 5) Star or Don't Know, 6) Merger.  The system receives back a classification, but they show the same galaxy to many users such that they get multiple classifications for a single galaxy. The result of the multiple classifications for each galaxy is is statistical certainty. Even trained astronomers make mistakes and disagree on the classification of some galaxies therefore having many amateurs classify a galaxy is better than a few professional astronomers. The system they have developed is also quite sensitive to user idiosyncrasy with respect to the fact that they monitor user performance for individual tasks such as ability to identify galaxy bulges or spiral arms and then weight that user's responses according to their accuracy and consistency for each task. The great thing about having people's eyes on the data is that unexpected and unique discoveries are made possible. Computers make errors not discoveries.  The fantastic support for the project is very encouraging. They did a survey of over 10,000 users to discover their motivation. The most important motivator across all age groups was:

I want to contribute to science

The Cosmos isn't strange people are strange

All those good willed people, what could possibly go wrong? The Galaxy Zoo published a paper on the Chiral correlation function of galaxy spins, that is they investigate if spiral galaxies have a tendency to spin clockwise or counterclockwise, and they discovered more about people than galaxies. If galaxies had a spin tendency it would quite simply undermine physics because chirality is a fundamental property of particles and cosmology. Indeed they found a tendency for galaxies to spin counterclockwise, but when they began to display the mirrored images of the galaxies (expecting to receive more clockwise responses this time) they found the users behavior to be inconsistent.  This indicates that either people are strange or the user interface of Galaxy Zoo is strange. They sum up the results in their abstract:

After establishing and correcting for a certain level of bias in our handedness results we find the winding sense of the galaxies to be consistent with statistical isotropy. In particular we find no significant dipole signal, and thus no evidence for overall preferred handedness of the Universe.

This may seem like an obvious result because it is intuitively correct, but it is important to verify observationally what seems intuitively correct and further previous studies had found evidence for non statistical isotropy.

Hanny's Voorwerp

The unique discovery of Hanny's Voorwerp was made possible only by the citizen scientists of the project. Hanny's Voorwerp is a green amoeba like blob next to a spiral galaxy that was discovered by a galaxy zoo volunteer, Hanny van Arkel, and hence the name of the object.  It appears green in some optical images because of bright emission lines that dominate in the SDDS g band. Spectral analyis has shown that it is a highly ionized region leading to the hypothesis that it is the result of a powerful transient outburst because whatever energized the blob is now gone. Researchers hypothesize that Hanny's Voorwerp is a quasar light echo:

Hanny’s Voorwerp, is bright in the SDSS g band due to unusually strong [OIII] 4959, 5007 emission lines. We present the results of the first targeted observations of the object in the optical, UV and X-ray, which show that the object contains highly ionized gas. Although the line ratios are similar to extended emission-line regions near luminous AGN, the source of this ionization is not apparent. The emission-line properties, and lack of x-ray emission from IC 2497, suggest either a highly obscured AGN with a novel geometry arranged to allow photoionization of the object but not the galaxy’s own circumnuclear gas, or, as we argue, the first detection of a quasar light echo. In this case, either the luminosity of the central source has decreased dramatically or else the obscuration in the system has increased within 10⁵ years. This object may thus represent the first direct probe of quasar history on these timescales.

Galaxy Zoo 2: Mergers

The result of users classifying galaxies as mergers were 3000 prime candidates with which they wanted to compare to simulations. The goal of Galaxy Zoo 2 is to understand cosmic mergers. They use the crowd source model again because this allows exploring the entire parameter space quickly whereas a machine may often find a local solution, but would be limited in knowledge of  unique solutions. Galaxy Zoo 2 presents you, the user, with a 3x3 grid of galaxies. At the center is a real image of a galaxy and surrounding it are 8 simulated galaxies. You must select which simulations match the real image best. If you don't like any of your options you can click a button at the top and you get a slot machine effect of 8 new galaxies.  It is very satisfying to demand random sets of galaxies and wait for a winner that matches what you are looking for. I found myself looking at upwards of 500 galaxies for each galaxy merger I classified which sounds absurd until you try it and see how easy it is. After selecting a few galaxies you can fine tune your selected simulations within parameter space and then select your final best simulation. The entire application runs the simluations locally on the user's machine in Java so the Zoo leverages everything a user has to offer from their computer's CPU to their brain.

Science Zoo

There are more Zoos in development. Moon Zoo is coming soon which will be like Galaxy Zoo for classifying features on the lunar surface using high resolution images from the Lunar Reconnaissance Orbiter Camera. Other fields will also be using the Zoo model, but surely astronomy offers the most exciting and beautiful possibilities. Astronomy is facing a flood of data soon with next generation projects like the Large Synoptic Telescope coming online in a few years. LSST will produce some 30 terabytes of data a night. This may be such a large amount of data that volunteers wont be able to sift through it all. In this case volunteers could provide the training sets for machine learning systems that could accurately classify data. The Zoo model will continue because people want to contribute to science.


References:

Anze Slosar, Kate Land, Steven Bamford, Chris Lintott, Dan Andreescu, Phil Murray, Robert Nichol, M. Jordan Raddick, Kevin Schawinski, Alex Szalay, Daniel Thomas, & Jan Vandenberg (2008). Galaxy Zoo: Chiral correlation function of galaxy spins MNRAS, 392 (1225) arXiv: 0809.0717v2

Chris Lintott, Kevin Schawinski, William Keel, Hanny van Arkel, Nicola Bennert, Edward Edmondson, Daniel Thomas, Daniel Smith, Peter Herbert, Matt Jarvis, Shanil Virani, Dan Andreescu, Steven Bamford, Kate Land, Phil Murray, Robert Nichol, Jordan Raddick, Anze Slosar, Alex Szalay, & Jan Vandenberg (2009). Galaxy Zoo : 'Hanny's Voorwerp', a quasar light echo? MNRAS arXiv: 0906.5304v1