Benchmarking for nonlinear system identification: submission platform and baseline results

Benchmarking identification methods is essential to validate newly proposed algorithms. The machine learning community has illustrated how benchmarks can be a driver for new developments in the field. However, thorough benchmarking is not commonplace in the system identification literature, and good benchmarking datasets and leaderboards are missing. This paper reflects on what constitutes a good system identification benchmark and introduces the nonlinearbenchmark.org platform for nonlinear system identification benchmarks and leaderboard together with a set of baseline results that have been generated for a set of widely used and promising benchmarking datasets.