IWAI2020 onlinesysid

Code and experiments for the paper entitled

"Online system identification in a Duffing oscillator by free energy minimisation" ,

presented at the International Workshop on Active Inference 2020.

Content

• FEM_prederror.ipynb and FEM_simerror.ipynb are Jupyter notebooks containing the method described in the paper. The first is a 1-step ahead prediction error experiment and the other a simulation error experiment. It uses ForneyLab.jl and a custom node called "NLARX" provided here (NLARX-node folder). If you don't have Jupyter installed, you can read the notebook by opening FEM_prederror.html or FEM_simerror.html in a browser.

• PEM_prederror.m and PEM_simerror.m are baseline methods implemented using Matlab's System Identification Toolbox. The trained model is stored in models/narx_sigmoidnet4.mat. Results can be loaded directly via results/results_narx_sigmoidnet4_ksteppred.mat or results/results_narx_sigmoidnet4_simulation.mat.

• Data comes from the Nonlinear Benchmark, specifically the Silverbox problem.

Contact

Problems with running code or feedback on the method can be given in the issues tracker.