Heuristic machinery to uncover hidden features of SU($N$) Fermions with neural networks.

The power of machine learning (ML) provides the possibility of analyzing experimental measurements with an unprecedented sensitivity. However, it still remains challenging to uncover hidden features directly related to physical observables and to understand physics behind from ordinary experimental data using ML. Here, we introduce a heuristic machinery by combining the power of ML and the "trial and error" in a supervised way. We use our machinery to reveal hidden thermodynamic features in the density profile of ultracold fermions interacting within SU($N$) spin symmetry prepared in a quantum simulator, and discover their connection to spin multiplicity. Although such spin symmetry should manifest itself in a many-body wavefuction, it is elusive how the momentum distribution of fermions, the most ordinary measurement, reveals the effect of spin symmetry. Using a fully trained convolutional neural network (NN) with a remarkably high accuracy of $\sim$94$\%$ for detection of the spin multiplicity, we investigate the dependency of accuracy on various hidden features with filtered measurements. Guided by our machinery, we verify how the NN extracts a thermodynamic compressibility from density fluctuations within the single image. Our machine learning framework shows a potential to validate theoretical descriptions of SU($N$) Fermi liquids, and to identify hidden features even for highly complex quantum matters with minimal prior understanding.