Approximate Dynamic Programming: An Efficient Machine Learning Algorithm

We propose an efficient machine learning algorithm for two-stage stochastic programs. This is termed as projected hybrid algorithm, and consists of sub-gradient piecewise linear approximation methods. use the sample information to update on objective function. Then we introduce a projection step, which implemented methods, jump out from local optimum, so that can achieve global optimum. By innovative show convergent property general Furthermore, network recourse problem, our drop steps, but still maintains convergence property. Thus, if properly construct initial functions, pure method The proposed approximate dynamic programming overcomes high dimensional state variables using methods learning, its logic capture critical ability structure anticipate impact decisions now future. optimization framework, carefully calibrated against historical performance, make it possible changes in collective intelligence experienced decisions. Computational results indicate exhibits rapid convergence.