Interval predictor model

In regression analysis, an interval predictor model (IPM) is an approach to regression where bounds on the function to be approximated are obtained. This differs from other techniques in machine learning, where usually one wishes to estimate point values or an entire probability distribution. Interval Predictor Models are sometimes referred to as a nonparametric regression technique, because a potentially infinite set of functions are contained by the IPM, and no specific distribution is implied for the regressed variables.

Multiple-input multiple-output IPMs for multi-point data commonly used to represent functions have been recently developed.[1] These IPM prescribe the parameters of the model as a path-connected, semi-algebraic set using sliced-normal [2] or sliced-exponential distributions.[3] A key advantage of this approach is its ability to characterize complex parameter dependencies to varying fidelity levels. This practice enables the analyst to adjust the desired level of conservatism in the prediction.

As a consequence of the theory of scenario optimization, in many cases rigorous predictions can be made regarding the performance of the model at test time.[4] Hence an interval predictor model can be seen as a guaranteed bound on quantile regression. Interval predictor models can also be seen as a way to prescribe the support of random predictor models, of which a Gaussian process is a specific case .[5]

Convex interval predictor models

edit

Typically the interval predictor model is created by specifying a parametric function, which is usually chosen to be the product of a parameter vector and a basis. Usually the basis is made up of polynomial features or a radial basis is sometimes used. Then a convex set is assigned to the parameter vector, and the size of the convex set is minimized such that every possible data point can be predicted by one possible value of the parameters. Ellipsoidal parameters sets were used by Campi (2009), which yield a convex optimization program to train the IPM.[4] Crespo (2016) proposed the use of a hyperrectangular parameter set, which results in a convenient, linear form for the bounds of the IPM.[6] Hence the IPM can be trained with a linear optimization program:

 

where the training data examples are   and  , and the Interval Predictor Model bounds   and   are parameterised by the parameter vector  . The reliability of such an IPM is obtained by noting that for a convex IPM the number of support constraints is less than the dimensionality of the trainable parameters, and hence the scenario approach can be applied.

Lacerda (2017) demonstrated that this approach can be extended to situations where the training data is interval valued rather than point valued.[7]

Non-convex interval predictor models

edit

In Campi (2015) a non-convex theory of scenario optimization was proposed.[8] This involves measuring the number of support constraints,  , for the Interval Predictor Model after training and hence making predictions about the reliability of the model. This enables non-convex IPMs to be created, such as a single layer neural network. Campi (2015) demonstrates that an algorithm where the scenario optimization program is only solved   times which can determine the reliability of the model at test time without a prior evaluation on a validation set.[8] This is achieved by solving the optimisation program

 

where the interval predictor model center line  , and the model width  . This results in an IPM which makes predictions with homoscedastic uncertainty.

Sadeghi (2019) demonstrates that the non-convex scenario approach from Campi (2015) can be extended to train deeper neural networks which predict intervals with hetreoscedastic uncertainty on datasets with imprecision.[9] This is achieved by proposing generalizations to the max-error loss function given by

 

which is equivalent to solving the optimisation program proposed by Campi (2015).

Applications

edit

Initially, scenario optimization was applied to robust control problems.[10]

Crespo (2015) and (2021) applied Interval Predictor Models to the design of space radiation shielding [11] and to system identification.[12]

In Patelli (2017), Faes (2019), and Crespo (2018), Interval Predictor models were applied to the structural reliability analysis problem.[13] [5] [14] Brandt (2017) applies interval predictor models to fatigue damage estimation of offshore wind turbines jacket substructures.[15]

Garatti (2019) proved that Chebyshev layers (i.e., the minimax layers around functions fitted by linear  -regression) belong to a particular class of Interval Predictor Models, for which the reliability is invariant with respect to the distribution of the data.[16]

Software implementations

edit

OpenCOSSAN provides a Matlab implementation of the work of Crespo (2015).[13]

References

edit
  1. ^ Crespo, Luis G.; Kenny, Sean P.; Colbert, Brendon K.; Slagel, Tanner (2021). "Interval Predictor Models for Robust System Identification". 2021 60th IEEE Conference on Decision and Control (CDC). pp. 872–879. doi:10.1109/CDC45484.2021.9683582. ISBN 978-1-6654-3659-5. S2CID 246479771.
  2. ^ Crespo, Luis; Colbert, Brendon; Kenny, Sean; Giesy, Daniel (2019). "On the quantification of aleatory and epistemic uncertainty using Sliced-Normal distributions". Systems and Control Letters. 34: 104560. doi:10.1016/j.sysconle.2019.104560. S2CID 209339118.
  3. ^ Crespo, Luis G.; Colbert, Brendon K.; Slager, Tanner; Kenny, Sean P. (2021). "Robust Estimation of Sliced-Exponential Distributions". 2021 60th IEEE Conference on Decision and Control (CDC). pp. 6742–6748. doi:10.1109/CDC45484.2021.9683584. ISBN 978-1-6654-3659-5. S2CID 246476974.
  4. ^ a b Campi, M.C.; Calafiore, G.; Garatti, S. (2009). "Interval predictor models: Identification and reliability". Automatica. 45 (2): 382–392. doi:10.1016/j.automatica.2008.09.004. ISSN 0005-1098.
  5. ^ a b Crespo, Luis G.; Kenny, Sean P.; Giesy, Daniel P. (2018). "Staircase predictor models for reliability and risk analysis". Structural Safety. 75: 35–44. doi:10.1016/j.strusafe.2018.05.002. ISSN 0167-4730. S2CID 126167977.
  6. ^ Crespo, Luis G.; Kenny, Sean P.; Giesy, Daniel P. (2016). "Interval Predictor Models With a Linear Parameter Dependency". Journal of Verification, Validation and Uncertainty Quantification. 1 (2): 021007. doi:10.1115/1.4032070. ISSN 2377-2158.
  7. ^ Lacerda, Marcio J.; Crespo, Luis G. (2017). "Interval predictor models for data with measurement uncertainty". 2017 American Control Conference (ACC). pp. 1487–1492. doi:10.23919/ACC.2017.7963163. hdl:2060/20170005690. ISBN 978-1-5090-5992-8. S2CID 3713493.
  8. ^ a b Campi, Marco C.; Garatti, Simone; Ramponi, Federico A. (2015). "Non-convex scenario optimization with application to system identification". 2015 54th IEEE Conference on Decision and Control (CDC). pp. 4023–4028. doi:10.1109/CDC.2015.7402845. ISBN 978-1-4799-7886-1. S2CID 127406.
  9. ^ Sadeghi, Jonathan C.; De Angelis, Marco; Patelli, Edoardo (2019). "Efficient Training of Interval Neural Networks for Imprecise Training Data". Neural Networks. 118: 338–351. doi:10.1016/j.neunet.2019.07.005. PMID 31369950. S2CID 199383010.
  10. ^ Campi, Marco C.; Garatti, Simone; Prandini, Maria (2009). "The scenario approach for systems and control design". Annual Reviews in Control. 33 (2): 149–157. doi:10.1016/j.arcontrol.2009.07.001. ISSN 1367-5788.
  11. ^ Crespo, Luis G.; Kenny, Sean P.; Giesy, Daniel P.; Norman, Ryan B.; Blattnig, Steve (2016). "Application of Interval Predictor Models to Space Radiation Shielding". 18th AIAA Non-Deterministic Approaches Conference. doi:10.2514/6.2016-0431. hdl:2060/20160007750. ISBN 978-1-62410-397-1. S2CID 124192684.
  12. ^ Crespo, Luis G.; Kenny, Sean P.; Colbert, Brendon K.; Slagel, Tanner (2021). "Interval Predictor Models for Robust System Identification". 2021 60th IEEE Conference on Decision and Control (CDC). pp. 872–879. doi:10.1109/CDC45484.2021.9683582. ISBN 978-1-6654-3659-5. S2CID 246479771.
  13. ^ a b Patelli, Edoardo; Broggi, Matteo; Tolo, Silvia; Sadeghi, Jonathan (2017). "Cossan Software: A Multidisciplinary and Collaborative Software for Uncertainty Quantification". Proceedings of the 2nd International Conference on Uncertainty Quantification in Computational Sciences and Engineering (UNCECOMP 2017). pp. 212–224. doi:10.7712/120217.5364.16982. ISBN 978-618-82844-4-9.
  14. ^ Faes, Matthias; Sadeghi, Jonathan; Broggi, Matteo; De Angelis, Marco; Patelli, Edoardo; Beer, Michael; Moens, David (2019). "On the robust estimation of small failure probabilities for strong non-linear models". ASCE-ASME Journal of Risk and Uncertainty in Engineering Systems, Part B: Mechanical Engineering. 5 (4). doi:10.1115/1.4044044. ISSN 2332-9017. S2CID 197472507.
  15. ^ Brandt, Sebastian; Broggi, Matteo; Hafele, Jan; Guillermo Gebhardt, Cristian; Rolfes, Raimund; Beer, Michael (2017). "Meta-models for fatigue damage estimation of offshore wind turbines jacket substructures". Procedia Engineering. 199: 1158–1163. doi:10.1016/j.proeng.2017.09.292. ISSN 1877-7058.
  16. ^ Garatti, S.; Campi, M.C.; Carè, A. (2019). "On a class of Interval Predictor Models with universal reliability". Automatica. 110: 108542. doi:10.1016/j.automatica.2019.108542. hdl:11311/1121161. ISSN 0005-1098. S2CID 204188183.