Virtual Classroom

Evaluation/Validation of Machine Learning Models - Remote training

MLEVAL-EN-D

Who should attend?

  • E&P engineers are used to work with models. Until recently all the models used in the oil and gas industry where models based on a relative small number of data and strong physical relationships to link the observations and to help the decision process by a better understanding of the subsurface.
  • In the 90’s first and more abundantly after 2012 some statistical models, just based on data and not on physical laws, have surfaced. They support already a lot of processes in the O&G Industry such as the mechanical failure prediction of rotary equipment, video surveillance analysis, seismic or well log interpretation, data mining… but it can be said that this models are usually "opaque".
  • Cooperation between these new type of data models and engineers depends on trust. If the engineers are to accept algorithm prescriptions, they need to trust them.
  • This training will present several ways to measure the quality of the models as well as some possibilities to anticipate their behavior on particular data set in order to evaluate if a model is trustable or not.
Public :
  • Oil and Gas engineers who have to interact with data scientist or who have to use trustable machine learning models to achieve their business objectives

Level :Awareness

Prerequisite :
  • No pre-requisistes are necessary to follow this course.

Course Content

  • INTRODUCTION

  • WHAT IS SUPERVISED TRAINING IN THE AI DOMAIN?

      • Learning from data.
      • Various algorithms for machine learning, pros and cons.
      • Deep learning.
      • Object detection.
  • BENCHMARKING

      • Selecting training, benchmarking and blind test data set. The train/split approach.
      • K-folds cross validation.
      • True and false positives, true and false negatives.
      • Multi classes confusion matrix.
      • Practical exercise.
  • BENCHMARKING INTERPRETATION

      • F1 score.
      • Other accuracy criterias.
      • ROC and AUC.
      • The case of regional detections.
  • EXPLAINABLE AI-XAI

      • Lime.
      • Eli5.
      • SHAP.

Learning Objectives

  • At the end of the training, the trainees will be able to:
  • Understand the supervised machine learning approach: test, validation and prediction.

Ways & Means

  • This course can be delivered using in-person lectures or virtual classrooms. Each training module contains lectures, hands-on practices and/or case studies.

More

Coordinator :IFP Training trainers (permanent or contracted) having a good expertise and/or experience of the related topics, trained to adult teaching methods, and whose competencies are kept up-to-date.

To French entities : IFP Training is referenced to DataDock ; you may contact your OPCO about potential funding. Please contact our disabled persons referent to check the accessibility of this training program : referent.handicap@ifptraining.com