What is Machine Learning?
Machine Learning is an application of Artificial Intelligence based on the concept that machines should be given access to data and learn specific tasks by themselves, without being explicitly programmed. Machine learning exploits computing systems that learn and predict from data. It is particularly useful when working with large datasets, as it detects patterns and predicts (and in some cases, recommends) outcomes. Its ability to learn from data and build an experience is opposed to deterministic approaches, which require user instructions and a human knowledge base.
Machine Learning and Geoscience
Machine learning applications for geoscience data have been in use for over 25 years. With the massive growth in petrotechnical data, they have now become a practical necessity.
Among available computing systems, Artificial Neural Networks (ANNs), inspired by biological neural networks, are among the ones most often adopted by the Oil & Gas industry to deal with the increasing volume of seismic and well data. Artificial Neural Network computing is the study of networks of adaptable nodes which learn to perform tasks based on data exposure and experience, generally without being programmed with any task-specific rules. They are designed to classify information in a similar fashion to the human brain. For example, they can be taught to recognize images and classify them according to their different elements.
Emerson is a pioneer in implementing advanced, proven and reliable machine learning solutions.
Our machine learning-based technology is able to describe the subsurface from large amounts of various types of data. This allows users to:
- Describe and explain an existing outcome
- Predict what will happen
- Provide recommendations for risk management and decision-making
This strategy allows the application of machine learning and predictive analytics to prospecting, field development and production optimization.
Machine learning methodologies may be divided into Unsupervised and Supervised Learning.
Input data is not labeled (soft data); the goal is to find similarities in the data point for grouping similar data points together. A model is prepared by deducing structures present in the input data. It may be done through a mathematical process to systematically reduce redundancy, or to organize data by similarity. Some of the algorithms include clustering, dimensionality reduction and association rule learning.
Examples of Emerson Unsupervised Machine Learning
|♦ Self-organizing mapping (SOM)
||♦ Gaussian processes
|♦ K-means cluster analysis
||♦ Back propagation neural network
|♦ Ascending hierarchical clustering
||♦ Self-growing neural network
|♦ Dynamic clustering
||♦ Principal component analysis (PCA)
|♦ Hybrid classification
||♦ Multi-resolution graph-based clustering
Input data has a known label (hard data). A model is prepared through a training process which requires predictions, and corrections are made when those predictions are wrong. The training process continues until the model achieves the desired level of accuracy on the training data. Problems include classification and regression.
Examples of Emerson Supervised Machine Learning
|♦ Neural network "ensembles" or Democratic Neural Network Association (DNNA)TM
||♦ Convolutional neural network
In recent years, the explosion in data size and quantity, the development of GPUs, and falling hardware prices have led to intensified research into advanced AI algorithms. A new class of Machine Learning techniques which can handle Big Data has emerged: Deep Learning.
Deep learning is a class of machine learning algorithms that:
- Use a cascade of multiple layers of nonlinear processing units for feature extraction and transformation. Each successive layer uses the output from the previous layer as input.
- Learn in supervised (e.g., classification) and/or unsupervised (e.g., pattern analysis) manners.
- Learn multiple levels of representations that correspond to different levels of abstraction.
Emerson solutions that use Machine Learning