Accuracy Assessment and Reinforcement Training of Deep Neural Networks in the Context of Accelerating a Simulation-Based Toolpath Generation Method

Despite the success of deep neural networks (DNNs) in many fields, their use in scientific domains remains challenging. DNNs give over-confident results and do not directly report a confidence measure alongside their output. The use of DNNs is thus often restricted to more error-tolerant domains. In scientific modeling, we usually understand the system’s structure and the process mechanism in-depth. It is, however, difficult to incorporate this knowledge into a DNN. Nevertheless, there is a growing interest in applying DNNs to scientific computing. This paper presents a novel method for integrating the knowledge from simulation-driven systems into DNNs. In particular, we illustrate the use of the method to accelerate a simulation-driven toolpath generation method. The proposed method has a regularization effect and can add the right inductive bias to the model. Where backtracking is possible, one can also use the method to assess the accuracy of the predictions and take corrective action. © 2025 U-turn Press LLC.