Active sample selection in scalar fields exhibiting non-stationary noise with parametric heteroscedastic Gaussian process regression

Wilson, Troy; Williams, Stefan B.

doi:10.1109/icra.2017.7989763

Cited by 3 publications

(3 citation statements)

References 20 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Heteroscedastic noise with parametric noise models can be learnt via maximum likelihood [10]. A Bayesian approach is to add a second GP prior to the log-variance of the noise model [21].…”

Section: Related Workmentioning

confidence: 99%

“…The noise model was computed using the regression model in (10). Something else to note about the GP hyper-parameters is that their values have to be proportional to the range of the expected cumulative reward function to model for standard comparison among functions, so the expected cumulative rewards were scaled to [0, 100].…”

Section: B Bo Hyper-parameter Search Space and Function Scalingmentioning

confidence: 99%

“…Noise in the highest rewards region is significantly less than elsewhere, evidencing an input-dependent behaviour known as heteroscedasticity. This behaviour has been approached in different ways in the BO literature [7], [10], leading to performance improvements.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Heteroscedastic Bayesian Optimisation for Stochastic Model Predictive Control

Guzman¹,

Oliveira²,

Ramos³

2020

Preprint

View full text Add to dashboard Cite

Model predictive control (MPC) has been successful in applications involving the control of complex physical systems. This class of controllers leverages the information provided by an approximate model of the system's dynamics to simulate the effect of control actions. MPC methods also present a few hyperparameters which may require a relatively expensive tuning process by demanding interactions with the physical system. Therefore, we investigate fine-tuning MPC methods in the context of stochastic MPC, which presents extra challenges due to the randomness of the controller's actions. In these scenarios, performance outcomes present noise, which is not homogeneous across the domain of possible hyper-parameter settings, but which varies in an input-dependent way. To address these issues, we propose a Bayesian optimisation framework that accounts for heteroscedastic noise to tune hyper-parameters in control problems. Empirical results on benchmark continuous control tasks and a physical robot support the proposed framework's suitability relative to baselines, which do not take heteroscedasticity into account.

show abstract

“…Heteroscedastic noise with parametric noise models can be learnt via maximum likelihood [10]. A Bayesian approach is to add a second GP prior to the log-variance of the noise model [21].…”

Section: Related Workmentioning

confidence: 99%

Section: B Bo Hyper-parameter Search Space and Function Scalingmentioning

confidence: 99%