In the last years, neural networks (NNs) have evolved from laboratory environments to the state-of-the-art for many real-world problems. Due to the properties of the loss surface and training methods, we hypothesize that NN models evolve on unique, smooth trajectories in weight space during training. Following, a population of such NN models (referred to as “model zoo”) would form topological structures in weight space. Further, the geometry, curvature and smoothness of these structures may con…