学术翻译：固定伪梯度样本数量导致算法线性收敛 | F纳什均衡

The academic translation is as follows:

If the number of samples of pseudo-gradients is fixed, it will no longer guarantee the progressive reduction of uncertainty, and therefore, it will no longer provide precise convergence in terms of expectation for the algorithm. The following inference indicates that when a constant sampling capacity is used in Algorithm 1, only linear convergence, i.e., F Nash equilibrium, can be achieved in the neighborhood of the optimal solution. It can be observed that the boundary D depends on the network structure, batch size, step size, and problem parameters η, L, and ν.