Fusion ignition in the laboratory was first demonstrated in indirect drive inertial confinement fusion experiments at the National Ignition Facility, marking a major scientific milestone, and has been achieved multiple times since. Now that igniting hot spots can be created experimentally, understanding hot spot properties is a key challenge in pushing for higher gain experiments. The conditions of the hot spot plasma, where fusion reactions first take place, are often inferred using simple physics models as they cannot be directly measured in experiments. The interpretability and past validation of these models make them an appealing choice, but the simplifying assumptions made about the hot spot are not always valid, particularly in higher yield experiments when hot spot conditions rapidly change. Here, we present two deep learning approaches to inferring hot spot conditions from experimentally measured quantities or “observables”; the first a “black box” model using an ensemble of neural networks and the second a symbolic regression model, using a recurrent neural network to learn a mathematical expression for a given hot spot parameter. A set of over 100,000 2D HYDRA simulations are used to train and test the deep learning models and results are compared with the 0D model. We find both deep learning models outperform the frequently used 0D analytical model across the simulation ensemble, and the learned symbolic expressions resemble the 0D model expressions with additional adjustments involving additional observables not found in the 0D model. We hypothesize physical motivations for adjustments and demonstrate symbolic regression methods have the potential to improve upon existing models by leveraging the capability of deep learning while maintaining the interpretability of physics models.