A Comprehensive Benchmark of Neural Networks for System Identification
Preprint
Abstract
This paper compares a wide variety of neural network architectures applied in the context of black-box modeling for robotics and control. We compare six different architectural concepts and four activation functions, with over three hundred different models. Those models were applied to three robotics datasets to show the differences in performance between the architectures along with their limitations.Keywords
All Related Versions
This publication has 4 references indexed in Scilit:
- Recurrent Neural Networks for Multivariate Time Series with Missing ValuesScientific Reports, 2018
- Information theoretic MPC for model-based reinforcement learning2017
- The EuRoC micro aerial vehicle datasetsThe International Journal of Robotics Research, 2016
- Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation2014