Stochastic mesh adaptive direct search for blackbox optimization using probabilistic estimates

Audet, Charles; Dzahini, Kwassi Joseph; Kokkolaras, Michael; Digabel, Sébastien Le

doi:10.1007/s10589-020-00249-0

Cited by 26 publications

(19 citation statements)

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In particular, we consider phase retrieval, which seeks to minimize the function, and blind deconvolution, which seeks to minimize Both of these applications are ones in which Common Random Numbers (CRNs) are a reasonable assumption, making two-point gradient estimates relevant. In particular, in (18), the pairs (a i , b i ) can be held constant between two function evaluations, and in (19), triplets (u i , v i , b i ) can be fixed as well.…”

Section: Numerical Resultsmentioning

confidence: 99%

“…These are nice examples of how good the zeroth order algorithm works when the stepsize is properly chosen. (18) min…”

Section: Comparison With Methods Using a Stochastic Subgradient Oraclementioning

confidence: 99%

“…From the plots it can be seen that our algorithm (with suitable choices of the stepsize) outperforms (20) We would like to notice that the MADS algorithm was originally developed for deterministic blackbox optimization. Recently a stochastic variant of this approach was proposed in [18].…”

Section: Comparison With a Naive Stochastic Variant Of Nomadmentioning

confidence: 99%

See 2 more Smart Citations

A zeroth order method for stochastic weakly convex optimization

Kungurtsev

Rinaldi

2021

Comput Optim Appl

View full text Add to dashboard Cite

In this paper, we consider stochastic weakly convex optimization problems, however without the existence of a stochastic subgradient oracle. We present a derivative free algorithm that uses a two point approximation for computing a gradient estimate of the smoothed function. We prove convergence at a similar rate as state of the art methods, however with a larger constant, and report some numerical results showing the effectiveness of the approach.

show abstract

Section: Numerical Resultsmentioning

confidence: 99%

“…These are nice examples of how good the zeroth order algorithm works when the stepsize is properly chosen. (18) min…”

Section: Comparison With Methods Using a Stochastic Subgradient Oraclementioning

confidence: 99%

See 1 more Smart Citation

A zeroth order method for stochastic weakly convex optimization

Kungurtsev

Rinaldi

2021

Comput Optim Appl

View full text Add to dashboard Cite

show abstract

“…This would require improving its performance on unimodal functions with high conditioning, and improving its performance at very early steps, for example by leveraging SMAC for creating the initial DoEs. Another important future work may include the extension of TREGO to the case of noisy observations, following recent results in DFO [47,48] and established BO techniques [49].…”

Section: Discussionmentioning

confidence: 99%

TREGO: a Trust-Region Framework for Efficient Global Optimization

Diouane¹,

Picheny²,

Riche³

et al. 2021

Preprint

View full text Add to dashboard Cite

Efficient Global Optimization (EGO) is the canonical form of Bayesian optimization that has been successfully applied to solve global optimization of expensive-to-evaluate black-box problems. However, EGO struggles to scale with dimension, and offers limited theoretical guarantees. In this work, we propose and analyze a trust-region-like EGO method (TREGO). TREGO alternates between regular EGO steps and local steps within a trust region. By following a classical scheme for the trust region (based on a sufficient decrease condition), we demonstrate that our algorithm enjoys strong global convergence properties, while departing from EGO only for a subset of optimization steps. Using extensive numerical experiments based on the well-known COCO benchmark, we first analyze the sensitivity of TREGO to its own parameters, then show that the resulting algorithm is consistently outperforming EGO and getting competitive with other state-of-the-art global optimization methods. The method is available both in the R package DiceOptim 1 and python library trieste 2 .

show abstract

“…These uncertainties arise from random initialization of the weight matrices, bias vectors, dropout, and descent step sizes during backpropagation. We are the first to use a stochastic derivative-free optimization algorithm to solve such a problem via the mesh adaptive direct search, known as StoMADS [21].…”

Section: Predictive Modelingmentioning

confidence: 99%

Predicting COVID-19 Incidences from Patients’ Viral Load using Deep-Learning

Khalil

Handawi

Chamseddine

et al. 2021

Preprint

Self Cite

View full text Add to dashboard Cite

The transmission of the contagious COVID-19 is known to be highly dependent on individual viral dynamics. Since the cycle threshold (Ct) isthe only semi-quantitative viral measurement that could reflect infectivity, we utilized Ct values to forecast COVID-19 incidences. Our COVID-19 cohort (n=9531), retrieved from a single representative cross-sectional virology test center in Lebanon, revealed that low daily mean Ct values are followed by an increase in the number of national positive COVID-19 cases. A subset of the data was used to develop a deep neural network model, tune its hyperparameters, and optimize the weights for minimal mean square error of prediction. The final model s accuracy is reported by comparing its predictions with an unseen dataset. Our model was the first to capture the interaction of the previously reported Ct values with the upcoming number of COVID-19 cases and any temporal effects that arise from population dynamics. Our model was deployed as a publicly available and easy-to-use estimator to facilitate prospective validation. Our model has potential application in predicting COVID-19 incidences in other countries and in assessing post-vaccination policies. Aside from emphasizing patient responsibility in adopting early testing practices, this study proposed and validated viral load measurement as a rigid input that can enhance outcomes and precision of viral disease predicting models.

show abstract

Stochastic mesh adaptive direct search for blackbox optimization using probabilistic estimates

Cited by 26 publications

References 32 publications

A zeroth order method for stochastic weakly convex optimization

A zeroth order method for stochastic weakly convex optimization

TREGO: a Trust-Region Framework for Efficient Global Optimization

Predicting COVID-19 Incidences from Patients’ Viral Load using Deep-Learning

Contact Info

Product

Resources

About