But anyone involved in the world of AI developments will recognize that we have become quite successful at tuning billions of parameters in neural networks, in part through a mathematical technique called gradient descent. And since this is central to training large neural networks, a lot of work has gone into optimizing the performance of gradient descent on GPUs.
So a team of researchers developed a process they call FLIP (a non-acronym for Physics-based Learning for Imaging Planets) that uses a physical model of the hardware (in this case, the NIRCam on the James Webb Space Telescope) at 106 possible parameters and then uses software to optimize all of its parameters. The optimization starts with a simulated point spread function and then uses gradient descent to optimize performance using data from an actual star. Running on a current GPU, this process only takes a little over half an hour.
The researchers then adjust the results based on the actual scientific picture they are analyzing. This final part of the process is crucial because some variables will differ between observations.
Good results
To test how this works, the researchers used observations of two stars with known exoplanets (HIP 65426 and 51 Eridani) as well as a third star (TYC 5899) that has no known exoplanets but is known to vary widely between observations. These are young stars, and two of them have massive planets that are still warm due to gravitational collapse, suggesting they have been imaged before.
Using Webb data, the team used FLIP to increase the signal-to-noise ratio by a factor of 1.5 to 2, depending on the observation. At a wavelength that can detect carbon dioxide, they were able to detect one of the planets in the Webb data for the first time. For TYC 5899, the approach increased the sensitivity of the observations by a factor of 15.
The researchers point out that the applications go beyond a better view of known planets. By reducing noise, it should also be possible to detect weaker (and therefore older or lower mass) specimens that had previously escaped detection. They also think they can modify FLIP to highlight where it makes the most dramatic changes to the model, which can potentially tell us useful information about what’s going on with the telescope’s hardware.
Nature, 2026. DOI: 10.1038/s41586-026-11213-z (About DOIs).
