Galaxies process the raw materials of the universe. Thus, studying the stars in galaxies and their histories gives us an insight into how the universe has changed with time, how galaxies likely formed, and how they will evolve. We commonly estimate the star formation histories of galaxies in a procedure called stellar population modeling. In this procedure, we traditionally compare sums of single age, single metallicity, simple stellar populations to the integrated spectra or photometry of an observed galaxy. In our work, we use the machine learning algorithm diffusion k-means to form a basis set of average simple stellar populations. Even with low signal-to-noise spectra, we show we can derive accurate star formation histories of galaxies more precisely than a traditional method. This method, however, only returns a single metallicity star formation history. Subsequent epochs of star formation are formed from the interstellar medium enriched by previous generations of star formation, so we expect an evolution in the metallicity of the stars in galaxies as galaxies evolve. To capture the complete (stellar age and metallicity) star formation history, we have begun to explore using diffusion k-means to form a basis set of multiple metallicities. In addition, I have begun work to fully characterizing today’s HgCdTe photodiode arrays to lay the foundation for future near infrared detector development. Low read noise and well-characterized detectors in the infrared increase the spectral information we collect from galaxies and are crucial in the emerging search for biosignatures, possible signs of life, in exoplanet atmospheres. Coincidently, using machine learning techniques, we see hints that in addition to advancing hardware, we will need to update how we analyze IR detector data to achieve cutting edge performance.