Additive manufacturing, and more specifically laser powder bed fusion (LPBF), complements conventional manufacturing by producing a low volume of highly complex functional metallic components. The mode of laser emission, either continuous (c-LPBF) or modulated (m-LPBF), has a pronounced impact on the resulting component. Although both have merits, the c-LPBF process dominates commercial machines and academic efforts. To promote further exploration of m-LPBF, which has the potential to reduce defects and enhance microstructural control, the process must strike a balance between component quality and the industry’s demand for increased production. In this work, a multi-objective optimization framework was adopted to balance time and quality of m-LPBF produced Ti-6Al-4V as a function of key processing parameters. Lacking an analytical model, Bayesian inference of Gaussian process regression was utilized to relate laser power, exposure time, point distance, and hatch spacing to the as-built relative density, serving as a proxy for quality, while batch active learning efficiently sampled the design space. In combination, this model accurately captured the relationship in a modest number of experiments and, in conjunction with the application of non-dominated sorting genetic algorithm II, was able to determine a non-dominated set of solutions approximating the Pareto front. Despite the model's accuracy, the current work highlights the need for a sufficiently large data set to accurately reflect the underlying mechanisms occurring in the m-LPBF process.