Neural Structured Learning (NSL) is a TensorFlow framework that improves neural network training by leveraging structured signals such as graphs and adversarial perturbations. It provides Keras APIs and low-level TF ops for incorporating explicit (graph) and implicit (adversarial) structure during training, along with tools for graph construction. The framework supports supervised, semi-supervised, and unsupervised learning scenarios and is designed to work with any neural network architecture including feed-forward, CNNs, and RNNs.