Page 9 - Machine Learning for Subsurface Characterization

Preface xxi

l Fast decision-making by processing high-speed data flow generated from
multiple sources/channels
l Develop data-driven models that improve over time to better represent the
physical processes, systems, and phenomena
l Facilitate precision engineering and characterization, especially for diagno-
sis and insight generation
l Intelligently automate mundane, repetitive, low-risk tasks

Machine learning–driven success stories
With widespread adoption of sensing and larger quantities of digital data within
reach, there are numerous potential applications for machine learning. When
humans feed well-structured data, the ML algorithms can extract patterns,
trends, and relationship to recommend the most appropriate procedure to
accomplish a task. Companies like Amazon and Netflix invest heavily to train
machine learning models to build robust recommendation engines for making
relevant and accurate predictions that match the user’s tastes/preferences.
Alphabet’s life sciences division Verily has been successful with assessment
of specific disease risk and its progression to facilitate preventive measures.
IBM Watson Health is discovering off-label uses of existing drugs, improving
chronic disease management, and providing drug safety evaluation. ML
improves speed, efficiency, and effectivity of discovery, as corroborated by
many drug discovery projects. AlphaGo is one of the feats of machine learning.
AlphaGo uses deep learning techniques that combine “supervised learning from
human expert games and reinforcement learning from games of self-play.”
Narrowly applied AI will be crucial in automation, preventive maintenance,
and rapid decision-making. Even if AI gets the right answer only 90% of the
time, the benefits of being able to instantly react to incoming data streams with
certain accuracy will be extremely valuable [4]. Human beings have limited
memory and cognitive resources for solving very complex problems. Addi-
tional help from ML models in terms of handling complexity will improve
human productivity and efficiency. ML-based decision-making is not affected
by human emotional and physical states. Compared with human specialist, ML
can process volumes of information from varied fields in a short amount of time
to help generate valuable insights and reliable forecast. In addition to this,
machine learning models after robust training are immediately scalable, with
the potential for simultaneous use by any number of users. When given contin-
uous supply of data and sufficient time, ML methods can concurrently improve
in both breadth and depth, unlike a human, for example, a ML interface can be
an expert in geology, geophysics, and geological engineering, simultaneously
[7]. ML can analyze large populations to identify large-scale patterns for devel-
oping holistic approaches. At the same time, ML methods can be also designed
to aggregate data for precision/personalized applications. ML will enable auto-
mation of analytical activities such as segmentation, optimization, and predic-
tive modeling. ML will help us reduce our efforts in repetitive, cumbersome,