Machine learning models increasingly augment human processes, either by performing repetitive tasks more quickly or by providing systematic insights that help put human knowledge into perspective. UC Berkeley astronomers were surprised to find that both happened after modeling gravitational microlensing events, leading to a new unified theory of the phenomenon.
Gravitational lensing occurs when light from distant stars and other stellar objects bends around a closer object directly between it and the observer, briefly giving a brighter – but distorted – view of the object at the top. further. Depending on how the light bends (and what we know about the distant object), we can also learn a lot about the star, planet, or system around which the light bends.
For example, a momentary spike in brightness suggests a planetary body transiting the line of sight, and this type of anomaly in the reading, called “degeneracy” for some reason, has been used to spot thousands of exoplanets.
Due to the limitations of their observation, it is difficult to quantify these events and objects beyond a handful of basics like their mass. And the degenerations are generally considered to fall under two possibilities: that the distant light has passed closer either to the star or the planet in a given system. Ambiguities are often reconciled with other observed data, such as the fact that we know by other means that the planet is too small to cause the scale of observed distortion.
Keming Zhang, a graduate student at UC Berkeley, was looking for a way to quickly analyze and categorize these lensing events because they appear in large numbers as we watch the sky more regularly and in greater detail. He and his colleagues trained a machine-learning model on data from known gravitational microlensing events with known causes and patterns, then released it onto a bunch of less well-quantified others.
The results were unexpected: in addition to deftly calculating when an observed event fell into one of two main types of degeneracy, he found many that did not.
“The previous two theories of degeneracy deal with cases where the background star appears to pass close to the foreground star or the foreground planet. The AI algorithm showed us hundreds examples of not only these two cases, but also situations where the star does not pass close to the star or the planet and cannot be explained by any of the previous theories,” Zhang said in a statement. Berkeley.
As a result – and after some persuasion, since a graduate student questioning established doctrine is tolerated but perhaps not encouraged – they ended up proposing a new “unified” theory of how the degeneracy of these observations can be explained, of which the two known theories were simply the most frequent cases.
Schematic showing a simulation of a 3-lens degeneracy solution.
They reviewed two dozen recent papers observing microlensing events and found that astronomers had misclassified what they saw as one type or the other when the new theory matched the data better than both.
“People were seeing these microlensing events, which were actually exhibiting this new degeneracy, but just didn’t realize it. It was really just machine learning looking at thousands of events where it became impossible to miss,” said Scott Gaudi, professor of astronomy at Ohio State University and co-author of the paper.
To be clear, the AI did not formulate or come up with the new theory – it was entirely up to human intellect. But without the systematic and reliable calculations of AI, it is likely that the simplified and less correct theory would have persisted for many years. Just as people learned to trust calculators and later computers, we are learning to trust certain AI models to produce interesting truth without preconceptions and assumptions – that is, if we don’t We haven’t simply encoded our own preconceptions and assumptions into them.
The new theory and the description of the process leading to it are described in an article published in the journal Nature Astronomy. This is probably not new to astronomers among our readership (it was a pre-print last year), but machine learning and general science buffs may cherish this interesting development.
Tech
