Surrogate emulators of high fidelity data sources are a fundamental component in any machining learning framework for materials science. They provide the necessary computational efficiency for many important downstream tasks, such as optimization in design. Importantly, useful surrogate models must be developed while accounting for two common characteristics of materials science problems: uncertainty and limited data. The Deep Material Network (DMN) is a physics-informed machine learning framework designed to address the second [1]. It is an emulator that predicts the homogenized material reponse of a specific microstructure. Most usefuly, it can stably extrapolate to predict non-linear material responses even though it is trained on only cheap linear (elastic) data. In this talk, we present the Variational DMN. With this variant, we extend the DMN to naturally account for uncertainty in its prediction. Specifically, it accounts for the aleatoric uncertainty that is often present in a material system due to underlying stochasticity in the microstructure. The VDMN‘s uncertainty prediction also extrapolates, allowing the VDMN to quantify uncertainty in both linear and nonlinear material responses without the need for nonlinear data. We present the algorithmic advances necessary for these changes. The developed algorithm automatically calibrates a heteroskedastic uncertainty prediction – in other words, where the uncertainty varies naturally depending on the input constitutive parameters of the base phases. We conclude by presenting a series of examples exploring the strengths and limitations of the VDMN as a tool for accelerated uncertainty quantification in materials science. REFERENCES (maximum 2 references) [1] Shin D. et al., Deep Material Network via a Quilting Strategy: Visualization for explainability and Recursive Training for Improved Accuracy, NPJ: Computational Materials, Vol. 128, 2023.