The modern cosmological model describes how tiny density fluctuations in the early Universe evolved into today's cosmic web of galaxies and dark matter. Though the model successfully describes much of our present-day large-scale structure, it also has unexplained tensions. For example, early- and late-time observations make conflicting measurements of sigma-8, a parameter that describes the clumpiness of the Universe. Large optical cosmological surveys are ushering in a golden age for data-driven cosmology that can address this tension: the Rubin Observatory and the Dark Energy Spectroscopic Instrument will soon provide exquisitely detailed maps of the sky, making it possible to explore cosmology's fingerprints on large-scale structure. These modern observations warrant being analyzed with equally modern data science methods, and this research program will support a team of scientists at Johns Hopkins University to develop understandable machine learning tools that can interpret upcoming surveys and address the sigma-8 tension. To engage the next generation in astronomy, this program will also support the development of play-based STEM lessons that teach early elementary school students about light, shadows, moon phases, and eclipses through stories and play. The kindergarten lesson plans that are developed through this program will be made publicly available to astronomers and educators.<br/> <br/>The decadal survey’s Pathways to Discovery identified the crucial role that machine learning (ML) could play in the next decade, leading to transformative discoveries from the decade’s rich, upcoming data sets. While ML has historically been touted as a black box that can generate order-of-magnitude improvements at the cost of interpretability, this does not need to be the case – modern techniques are making it possible to develop ML tools that improve results while still being understandable and leading to physical discoveries. This research program will develop a toolkit of understandable ML methods for interpreting detailed optical galaxy surveys from the Rubin Observatory and the Dark Energy Spectroscopic Instrument to explore the sigma-8 tension. This research will 1) produce a statistical census of cosmological information at small scales, probing techniques for describing how sub-Mpc structures correlate with the underlying cosmological model, 2) produce a low-scatter galaxy cluster dynamical mass proxy using symbolic regression to provide a closed-form, complementary framework for quantifying the abundance of massive clusters at low redshift, and 3) develop a deep learning approach for estimating galaxy cluster ellipticity, a major source of systematic error in weak lensing cosmological analyses.<br/><br/>This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.