神经网络训练与非可微目标函数（英文版）

英文标题：Neural Network Training and Non-Differentiable Objective Functions中文摘要：本论文探讨了计算机视觉领域中非可微目标函数的训练问题，提出了使用可微的代理函数进行训练的解决方案，使得神经网络训练更易扩展到新型任务，包括可分解和不可分解的评估指标。英文摘要：Many important computer vision tasks are naturally formulated to have anon-differentiable objective. Therefore, the standard, dominant trainingprocedure of a neural network is not applicable since back-propagation requiresthe gradients of the objective with respect to the output of the model. Mostdeep learning methods side-step the pr