Minimum-Variance Control Allocation Considering Parametric Model Uncertainty

Grauer, Jared A.; Pei, Jian

doi:10.2514/6.2022-0749

“…We believed such a requirement (i.e., having a tighter bound to the target risk estimator) is important in a transfer learning scenario where the goal is to use a source model that can reduce the risk on the target datasets. Our outcome in this experiment confirms the assessment in the work of Grauer and Pei (2022), where they noticed that when model uncertainty is known and distributed among the variance, the performance, and reliability of the model are improved. Hence, by combining the idea of important sampling with the control variates technique to provide an accurate estimate for the Bayesian LMMSE estimator, we were able to estimate the error on the target tasks and equally distribute the variance error.…”

Section: Discussionsupporting

confidence: 89%

Improving Model Selection in Deep Supervised Transfer Learning Under Homogeneous Setting

Omondiagbe¹,

MacDonell²,

Licorish³

2022

Preprint

0

View full text Add to dashboard Cite

In traditional machine learning environments, the use of non-parametric error estimation to set the discriminative threshold of a classifier to achieve the best accuracy is very effective. This method is not effective in a transfer learning environment because it is only reliable when both the training and testing data have similar distributions which is not the case in a transfer learning setting. Although the use of control variate techniques has been proposed to exploit the information about the error in the training sample to reduce the error in the test sample, this method yields a finite variance and the model uncertainty is not distributed among the variance. In this paper, we proposed and test a new transfer learning validation method called control linear minimum mean-squared error (CLMMSE) for source model selection under homogeneous transfer learning settings with the absence of an adequate pre-trained source model. Our approach adopts the Bayesian linear minimum mean-squared error (LMMSE) and integrates the idea of importance sampling into a control variate approach to provide an accurate estimate for the LMMSE that is then used to select the optimal source model. By combining importance sampling with the control variate technique to reduce further the variance, we can achieve a much tighter bound with the LMMSE. This approach reduces the risk in the target domain under data shift. Experimental results on synthetic data under two data shift settings demonstrate the efficacy of our approach. A further experiment on two real-world datasets shows that we were able to improve the accuracy of two state-of-the-art models tested; Bert (0.94\% to 65\%) and CodeBERT (1.82\% to 18.2\%) when compared to using previous selection methods.

show abstract

“…In the feature space, Long et al (2018) demonstrated that after feature adaptation, the distribution divergence is greatly reduced while the input data remains the same. Although the DEV method was seen to outperform the IWCV method in an unsupervised setting by using a control variate in lowering the variance and working in the feature space, their method failed to model uncertainty under data and label distribution, which could introduce further bias (Grauer and Pei, 2022).…”

Section: Parametric Error Estimationmentioning

confidence: 99%

“…Further study has shown that when model uncertainties are known and distributed among the variance, the performance and reliability of the model is greatly improved (Grauer and Pei, 2022). A Bayesian minimum mean square error estimator (BEE) (Dalton and Dougherty, 2013a,b) was proposed, in which the expected true error is computed based on the posterior of the model parameters.…”

Section: Parametric Error Estimationmentioning

confidence: 99%

See 1 more Smart Citation

Improving Model Selection in Deep Supervised Transfer Learning Under Homogeneous Setting

Omondiagbe¹,

MacDonell²,

Licorish³

2022

Preprint

0

View full text Add to dashboard Cite

In traditional machine learning environments, the use of non-parametric error estimation to set the discriminative threshold of a classifier to achieve the best accuracy is very effective. This method is not effective in a transfer learning environment because it is only reliable when both the training and testing data have similar distributions which is not the case in a transfer learning setting. Although the use of control variate techniques has been proposed to exploit the information about the error in the training sample to reduce the error in the test sample, this method yields a finite variance and the model uncertainty is not distributed among the variance. In this paper, we proposed and test a new transfer learning validation method called control linear minimum mean-squared error (CLMMSE) for source model selection under homogeneous transfer learning settings with the absence of an adequate pre-trained source model. Our approach adopts the Bayesian linear minimum mean-squared error (LMMSE) and integrates the idea of importance sampling into a control variate approach to provide an accurate estimate for the LMMSE that is then used to select the optimal source model. By combining importance sampling with the control variate technique to reduce further the variance, we can achieve a much tighter bound with the LMMSE. This approach reduces the risk in the target domain under data shift. Experimental results on synthetic data under two data shift settings demonstrate the efficacy of our approach. A further experiment on two real-world datasets shows that we were able to improve the accuracy of two state-of-the-art models tested; Bert (0.94\% to 65\%) and CodeBERT (1.82\% to 18.2\%) when compared to using previous selection methods.

show abstract

Adaptive multi-objective control allocation with online actuator selection for over-actuated systems

Jafari

¹

,

Khaki‐Sedigh

²

,

Birk

³

2022

Int. J. Dynam. Control

0

View full text Add to dashboard Cite

Minimum-Variance Control Allocation Considering Parametric Model Uncertainty

Cited by 5 publications

References 28 publications

Improving Model Selection in Deep Supervised Transfer Learning Under Homogeneous Setting

Improving Model Selection in Deep Supervised Transfer Learning Under Homogeneous Setting

Improving Model Selection in Deep Supervised Transfer Learning Under Homogeneous Setting

Adaptive multi-objective control allocation with online actuator selection for over-actuated systems

Contact Info

Product

Resources

About