report of learning optimization

Current problem:

design algorithms is a laborious process.

Focus field:

automating the design of unconstrained continuous optimization algorithms.

Assumption:

1. undiscounted settting y=1
2. restrictthe dependence of π on the objective function f to objective values and gradients evaluated at current and past locations.

Methods put foward:

optimization based on reinforcement learning.

Process:

Learning:

The best combination in the group is far superior to one of the best, which means it will be better than the algorithm it can choose

Autonomous optimizer is consist of other algorithms,but better than any other optimization algorithm .

Advantages:

1. minimizes the amount of a priori assumptions made about objective functions and can instead take full advantage of the information about the actual objective functions of interest.
2. has no hyperparameters that need to be tuned by the user.

Disadvantage:

1. unconstrained continuous optimization algorithms.
2. it may be used to solve various common classes of optimization problem.

Reference:

paper source: https://arxiv.org/pdf/1606.01885.pdf

reinforcement introduction: https://www.cnblogs.com/NaughtyBaby/p/5438013.html