Polynomial networks, recognized for their ability to model system dynamics accurately and provide interpretable symbolic equations, benefit from the Tucker decomposition’s generalization over canonical polyadic decomposition (CPD), offering enhanced control and expressivity. The study evaluates TuckerNet’s performance, comparing it against CPD-based networks in learning functions and neural ODEs within complex systems. The findings demonstrate TuckerNet’s potential as a superior alternative for polynomial network construction, particularly in parameter-constrained models, while also highlighting aspects beyond decomposition that impact learning outcomes.