TY - CONF
T1 - DECONSTRUCTING THE INDUCTIVE BIASES OF HAMILTONIAN NEURAL NETWORKS
AU - Gruver, Nate
AU - Finzi, Marc
AU - Stanton, Samuel
AU - Wilson, Andrew Gordon
N1 - Funding Information:
The authors would like to thank Greg Benton and Pavel Izmailov for helpful comments on our original submission. This research is supported by an Amazon Research Award, Facebook Research, Google Research, NSF I-DISRE 193471, NIH R01DA048764-01A1, NSF IIS-1910266, and NSF 1922658 NRT-HDR: FUTURE Foundations, Translation, and Responsibility for Data Science.
Publisher Copyright:
© 2022 ICLR 2022 - 10th International Conference on Learning Representationss. All rights reserved.
PY - 2022
Y1 - 2022
N2 - Physics-inspired neural networks (NNs), such as Hamiltonian or Lagrangian NNs, dramatically outperform other learned dynamics models by leveraging strong inductive biases. These models, however, are challenging to apply to many real world systems, such as those that don't conserve energy or contain contacts, a common setting for robotics and reinforcement learning. In this paper, we examine the inductive biases that make physics-inspired models successful in practice. We show that, contrary to conventional wisdom, the improved generalization of HNNs is the result of modeling acceleration directly and avoiding artificial complexity from the coordinate system, rather than symplectic structure or energy conservation. We show that by relaxing the inductive biases of these models, we can match or exceed performance on energy-conserving systems while dramatically improving performance on practical, non-conservative systems. We extend this approach to constructing transition models for common Mujoco environments, showing that our model can appropriately balance inductive biases with the flexibility required for model-based control.
AB - Physics-inspired neural networks (NNs), such as Hamiltonian or Lagrangian NNs, dramatically outperform other learned dynamics models by leveraging strong inductive biases. These models, however, are challenging to apply to many real world systems, such as those that don't conserve energy or contain contacts, a common setting for robotics and reinforcement learning. In this paper, we examine the inductive biases that make physics-inspired models successful in practice. We show that, contrary to conventional wisdom, the improved generalization of HNNs is the result of modeling acceleration directly and avoiding artificial complexity from the coordinate system, rather than symplectic structure or energy conservation. We show that by relaxing the inductive biases of these models, we can match or exceed performance on energy-conserving systems while dramatically improving performance on practical, non-conservative systems. We extend this approach to constructing transition models for common Mujoco environments, showing that our model can appropriately balance inductive biases with the flexibility required for model-based control.
UR - http://www.scopus.com/inward/record.url?scp=85150338591&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85150338591&partnerID=8YFLogxK
M3 - Paper
AN - SCOPUS:85150338591
T2 - 10th International Conference on Learning Representations, ICLR 2022
Y2 - 25 April 2022 through 29 April 2022
ER -