I am comparing different machine learning techniques for learning dynamical systems (e.g. a system of ordinary differential equations), and so far I've used Long-Short-Term Memory Networks (LSTM) and other variations of Recurrent Neural Networks, Dynamic Bayesian Networks, and Symbolic Regression.
However, I know only a part of this fascinating domain, so I wanted to ask the community: Can you suggest other state-of-the-art machine learning techniques for learning dynamical systems? Black-box or white-box, it's not important; I am more focused on getting good data fitting for my application.
Thanks in advance for any suggestion :-)
There is feedforward neural network (FNN), HMM, etc.
Look at this tutorial as well - https://arxiv.org/ftp/arxiv/papers/1905/1905.07727.pdf
I have recently explored GANs (Generative adversarial networks) for similar tasks.
give it a try
I haven't experiment all of them, but there are some papers working on reinforcement learning in non-stationary environment that may interest you :
Article Learning driving styles for autonomous vehicles from demonstration
Article A Survey on Multi-Task Learning
Article Multi-Agent Actor-Critic for Mixed Cooperative-Competitive E...
have a nice day.
I published a paper entitled:
Learning the behavior of dynamical systems from examples
In 1989, not sure whether this is state of the art ;-)
There are some open source library that work in this direction
here an example https://github.com/lululxvi/deepxde
- Badges
- Science topic