This project aims to develop an interpretable symbolic model of material yield surfaces using symbolic regression. A yield surface defines the boundary in stress space at which a material transitions from elastic behavior to yielding and permanent deformation. In three-dimensional mechanics, this surface is represented in a six-dimensional stress space, which makes its identification and interpretation challenging. The motivation of the project is to obtain compact analytical expressions for yield surfaces that are both physically meaningful and useful in downstream computational mechanics tasks. The proposed approach uses PySR, a Python library for symbolic regression based on evolutionary search and genetic algorithms, to discover candidate analytical expressions from sampled yield-surface data. Unlike standard regression problems, this task involves modeling an implicit function: the available data consist of points lying on the yield surface itself, rather than input-output pairs with a dependent target variable. This makes the problem fundamentally different from conventional explicit symbolic regression. To address this challenge, the project introduces a reformulation of the problem together with a data augmentation strategy. The augmentation procedure is designed to make the implicit modeling problem tractable while also helping enforce important constraints on the learned representation. In particular, the approach supports both numerical and soft-constraint-based ways of guiding the regression process, improving the stability of the search and the physical plausibility of the resulting expressions. The project will investigate how these augmented datasets and constraint mechanisms affect the ability of PySR to recover interpretable and accurate symbolic descriptions of the yield surface. The longer-term objective is to provide interpretable constitutive models that can serve as efficient heuristics in computational material modeling and simulation. Compared with purely black-box models, symbolic expressions can offer greater transparency, easier physical interpretation, and potentially lower computational cost when embedded into simulation workflows. This may contribute to more accurate and more efficient simulation of material behavior, especially in applications where repeated evaluation of constitutive laws is required. The project is exploratory and methodological in nature, combining ideas from symbolic regression, implicit function modeling, constrained learning, and computational mechanics. The requested computational resources will be used to run large-scale symbolic regression searches, evaluate many candidate expressions, and compare reformulation and augmentation strategies across yield-surface datasets. Main supervisor: Moa Johansson, Chalmers University of Technology