Just about any field one can think of—finance, , national , science, medicine, sports, and many others—can utilize simulation analysis to evaluate risks and explore new possibilities.

The quality of such analysis depends on how the simulated environment is.

This blog post uses a Wasserstein GAN + gradient penalty to create lifelike financial . Two features lead to greatly improved results:

  • Using networks. RNNs can produce a correlation structure in the generated data that fully connected networks weren’t able to accomplish.

  • Providing the generator and discriminator with that tells them which environment they are operating in when evaluating a sample.

The additional information, in this case it’s a market volatility index, also offers a major benefit when we want to use the generator for a practical purpose. We can create different types of data depending on what kind of setting we tell the generator it’s working in.

https://hergott.github.io/gan_rnn/



Source link
thanks you RSS link
( https://www.reddit.com/r//comments/9e5vo2/r_gans_for_simulation_analysis_creating_realistic/)

LEAVE A REPLY

Please enter your comment!
Please enter your name here