NeurASP: Embracing Neural Networks into Answer Set Programming

Reasoning way to integrate sub-symbolic and symbolic can help identify perception mistakes that violate semantic computation. We demonstrate how NeurASP can constraints, which in turn can make perception more make use of a pre-trained neural network in symbolic robust. For example, a neural network for object detection computation and how it can improve the neural may return a bounding box and its classification "car," but it network's perception result by applying symbolic may not be clear whether it is a real car or a toy car. The reasoning in answer set programming. Also, distinction can be made by applying reasoning about the relations NeurASP can be used to train a neural network with the surrounding objects and using commonsense better by training with ASP rules so that a neural knowledge. Or when it is unclear whether a round object attached network not only learns from implicit correlations to the car is a wheel or a doughnut, the reasoner could from the data but also from the explicit complex conclude that it is more likely to be a wheel by applying commonsense semantic constraints expressed by the rules.