**2019-05-24**

Recurrent neural network are a promising neural network layout which provides a better performance than a normal topology, which is a three layer forward network. The main problem with recurrent neural network is to adjust the weights. Instead of testing out random weights with a brute force algorithm, the slightly faster idea is to use genetic algorithms for identifying the parameters more efficient.^{[1]}

A practical example which includes the pendulum problem was realized in the Java programming language.^{[2]} The problem with genetic algorithms is, that they can only solve simple problem, but are not useful for tackling complex problems like realtime-strategy games or robotics control application. This is the reason, why in most papers only toy problems like the inverted pendulum problem or the mountain cart are provided.

## References

- ↑ Jonathan A. Michaels: geneticRNN: A simple genetic training algorithm for recurrent neural networks, 2018, https://github.com/JonathanAMichaels/geneticRNN
- ↑ rootmolloch: Recurrent Neural Network with genetic backtracking, https://github.com/rootmolloch/SparseLocalBacktrackingSelfGenetics