Multi-Label Classification Neural Networks with Hard Logical Constraints

Multi-label classification (MC) is a standard machine learning problem in which data point can be associated with set of classes. A more challenging scenario given by hierarchical multi-label (HMC) problems, every prediction must satisfy hard constraints expressing subclass relationships between In this paper, we propose C-HMCNN(h), novel approach for solving HMC which, network h the underlying MC problem, exploits hierarchy information order to produce predictions coherent and improve performance. Furthermore, extend logic used express able specify complex relations among classes new model CCN(h), extends C-HMCNN(h) again exploit We conduct an extensive experimental analysis showing superior performance both CCN(h) when compared state-of-the-art models general setting logical constraints.