写作常用语句总结

Abstract

• For both the problems, the proposed approach is found to yield better accuracy compared to conventional residual based PINN algorithms.

Introduction

• state-of-the-art attacks 最先进的攻击
• We design and analyze XX, which enjoys dual advantages of XX andXX
• require XX times more than
• in scenarios where explicit expressions for the gradients are difficult
  or infeasible to obtain 在XX的情况下
• Spurred by that，this paper XX 受这刺激，这篇文章XX
• demonstrate the superior performance of XX for XX 我们展示了对于XX的XX方法的优越性能
• benchmark 基准
• be elaborated on 将被详细说明
  We call the ZO variant of signSGD ‘ZO-signSGD’, which will be elaborated on in Sec. 3.
• concretely 更具体的
• comparable or even better 相比，甚至更好
• two-fold benefit 两方面好处
• To investigate the XX performance on XX, we first apply the PPINN to solve the XX, and subxxsequently we apply the XX to solve XX
• have shown promising results for a series of nonlinear benchmark problems

原理

• The rationale behind XX is that XX
  eg:
  [1] The rationale behind the ZO gradient estimate is that although it is a biased approximation to the true gradient of f, it is unbiased to the gradient of the randomized smoothing version of f with parameter µ
• For illustrating the performance of the proposed approach, benchmark results using Monte Carlo simulation (MCS) [62] are generated
  due to XX由于
• this is likely because (推测)
• schematic of(描述xx的原理图)

实验

• be initialized by以什么初始设定

• 实验网络设置
  For solving the problem using the proposed approach, the unknown response u is first represented by using an FC- DNN with 2 inputs, 5 hidden layers and 50 neurons per hidden layer. The 2 inputs to the DNN are time t and decay parameter Z. Hyperbolic tangent (tanh) activation function is considered for all but the last layer. For the last layer, linear activation function is considered. The initial conditions in Eq. (22) is automatically satisfied by modifying the DNN output, uNN using Eq. (16), where ub = 1.0 and B = t,