Page 8 - Machine Learning for Subsurface Characterization

xx Preface

model development requires a high-performance computing to process huge
data volumes at a reasonable speed. DL is a powerful pattern recognition
method, but the DL approach severely limits explainability of its outcomes
and interpretability of the DL models. DL heavily relies on data annotation/
labeling quality. ML/DL method tends to be impressive when considering its
statistical performance over many samples, but they can be highly erroneous
in individual cases.
Over the last 7 years, incredible advances in machine learning have been
made with the advent of deep neural networks that are trained on “‘Big Data”
using very fast GPUs. These advances have benefited from the accumulation of
digitized data and ubiquitous deployment of robust sensor systems. In addition,
there is a wealth of openly available technologies that make it simpler and
cheaper to build and run machine learning algorithms. Many of the tools are
easily accessible and inexpensive, for example, public clouds like Microsoft
Azure and Amazon Web Services, allowing massive data crunching exercises
without the need to buy tons of hardware [3]. These advances have led to the
state of the art in computer vision and speech recognition, such that machines
have now exceeded the powers of human sensory perception in certain areas [4].
Machine learning has ushered a whole new way of doing business by pro-
pelling progress in automation, sensor-based industrial monitoring, and algo-
rithmic analysis of business processes. Now, computers can learn the tasks to
assist humans rather than merely doing as they’re told. AI as a research area
has been around in computer science since the 1950s (including its subfields
such as machine learning and deep learning). Recent boom in AI implementa-
tions and its popularity has been due to better algorithms leading to improved
accuracy, faster GPUs providing large compute power, large datasets for train-
ing the ML and DL algorithms, easily accessible ML platforms for developing
data-driven models, and cloud services providing easier access to computa-
tional resources [5].
Due to the proliferation of data and rapid advances in the predictive analyt-
ics, machine learning is attracting large financial investments. Venture capital-
ists funded 1028 AI-related startups last year. Technical conferences and
workshops promising to explain AI and demonstrate the power of AI have
become a common and widespread trend. The annual meeting of the World
Economic Forum in Davos this year included close to 10 panels related to
AI, for example, “Designing Your AI Strategy” and “Setting Rules for the
AI Race” [6]. Any technology advancing at a fast pace and with such breathless
enthusiasm should be brought under a thorough reality check.
Here are few tasks related to O&G upstream exploration and production that
are suitable for ML/DL implementations:

l Detecting minute changes, variations, and patterns in high-dimensional
datasets
l Finding similarity and dissimilarity among systems/processes at a
granular level