Modern machine learning techniques open doors to powerful data analysis strategies that were not feasible even a few years ago — approaches that could enable transformative discoveries with current and upcoming cosmological survey data. But can we trust the black box? In this talk, I will highlight both the opportunities and challenges in developing credible machine learning methods to interpret cosmological observations. I will focus on interpretability and domain adaptation as keystones for building machine learning models that yield trustworthy, physically meaningful results. Finally, I will show how machine learning can serve not merely as a tool for achieving “better” results at the expense of understanding, but as a scientific partner that can point us toward physical discovery.

Speaker: Michelle Ntampaka (STScI)

The 2025 Fall Colloquium talks are held on Wednesdays at 3:00 PM. This colloquium is hosted by STScI and will be held as an in-person and virtual event.

You may join in person at STScI’s John N. Bahcall Auditorium or virtually on the STScI Research YouTube channel.

Please direct questions or comments to contact above. The 2025 Fall Colloquium Committee members are: Nimisha Kumari (STScI), Elena Manjavacas (STScI), Jack Neustadt (JHU), Kevin Schlaufman (JHU), Adam Smercina (STScI), Ethan Vishniac (JHU).