Lead two Machine Learning R&D projects:Deep Reinforcement Learning (DRL) algorithm to control and coordinate a behind-the-meter distributed energy resources (DERs) system that involves: on-site Photovoltaic electricity production and Battery storage.• Developed the DRL algorithms and the training framework that is based on simulation models that approximate the considered DER system.Extraction of buildings characteristics using Unstructured Data (e.g., satellite imagery, drone optical and thermal imagery, Lidar data) and Machine Learning:• Developed a toolbox to extract building footprint from satellite imagery using Deep Learning semantic segmentation.• Work with computer vision scientists to develop algorithms for extracting 3D geometrical characteristics and automatic detection of thermal anomalies. Developed an algorithm based on gradient boosting machine and cross validation to generate prediction intervals

Work with commercial building domain scientists, engineers and stakeholders to bring statistical expertise to the analysis and reporting of data related critical research issues. Design and develop new statistical and machine learning tools to resolve the automated measurement and verification research questions, including, but not limited to, commercial building’s energy consumption prediction, energy savings uncertainty quantification, time series change points detection. Contribute to the future direction of data collections, data processing and data analysis in the field of measurement and verification.

Developed a model for predicting commercial building’s energy consumption using decision-tree ensemble methods. Performed a statistical analysis that cross-matched building permit data with energy use data to provide information on the energy savings that may be associated with di erent building retrofit. Designed a method based on machine learning tools to estimate the US electricity marginal prices using open data. Improved a statistical method to estimate sales volume from sales rank data for consumer appliances.

Developed a joint modeling of mean and variance method (heteroscedastic regression) to approximate stochastic computer codes. This work will be incorporated into uncertainty analysis IFP Energies Nouvelles commercial software.Improved a method to build a latin hyper-cube design with correlated parameters (using copula). This work was incorporated into uncertainty analysis IFP Energies Nouvelles commercial software.Developed a screening method using derivative-based sensitivity measures in order to detect the non-inuential inputs for an acceptable computational cost in the framework of high dimensional reservoir simulator.Participated in development of uncertainty analysis methodology capable of helping in choosing enhanced oil recovery (EOR) products.Evaluated performance of different optimization algorithms for history matching problems (inverse problems).

Developed a multivariate wavelet kernel regression method to approximate computer codes with highly nonlinear or discontinuous inputs/output relation.Developed a methodology to perform uncertainty and sensitivity analysis of computer models with time series outputs. This functional approximation method is based on wavelet decomposition and smoothing spline estimation.Employed a smoothing spline estimation method based on analysis of variance (ANOVA) expansion to approximate high-dimensional computer codes.Developed an iterative thresholding algorithm to solve the nonnegative garrote optimization problem.Designed and implemented a sensitivity analysis method. This work was incorporated into uncertainty analysis IFP Energies Nouvelles commercial software.Participated in development and test of various algorithms incorporated into uncertainty analysis IFP Energies Nouvelles commercial software.