436
is Deep Neural Networks. Neural networks that mimic the information flow in a
brain are increasingly getting attention from the computer science research community. Originally proposed in the late twentieth century, research on neural networks stagnated due to the limitation of the hardware at that time. The increased
computational power by GPUs with thousands of cores as well as in memory
matrix manipulations, re-started the neural network research again. Especially in
the computer vision domain, Deep Neural Networks (i.e., deep learning) proved
to be promising.
Some recent applications of deep learning in food–energy–water research
includes identification of objects (trees, farm fields, etc.) from remote sensing imagery, identification of healthy/unhealthy animals from the camera feed in farms as
well as automated detection of ripe fruits from trees. This research is evolving, and
in the future, it is expected to have more use cases for GPU usage for deep learning
purposes. Readers are encouraged to read on these tools but beware that these tools
are often a black box that affects their interpretability.
16.4 Computer Science as a Scientific Tool
The use of computer science in scientific methods is twofold:
Causal and mechanistic models often start with model building and theories. Then
these models are numerically solved and backed by simulations. Many such
models require a complicated process of solving differential equations that are
hard to solve by hand and paper. In addition, often the simulations are hard without the use of a computer. For example, SWAT and GCM techniques require
heavy use of supercomputers.
Data-driven/empirical models involve controlled experiments and statistical analysis. Often, these methods start with hypothesis creation followed by hypothesis
testing as well as hypothesis ranking. Since such approaches are hard to prove by
mathematical theories (due to uncertainties, complications, etc.), they often rely
on a large amount of data in hand (fourth Paradigm of Science). Since these big
datasets require extensive analysis, computational tools that use distributed data
analysis (e.g., big data tools such as Hadoop and Spark) are required to perform
those. For example, the ozone hole over Antarctica was detected by such an
approach.
16.4.1 Nexus Data from Computer Science Perspective
In order to understand the role of computer science in food, energy, and water
research, one may need to understand the different types of data that are related to
nexus research. Since the details of the importance of Data is explained in Chap. 14,
E. Eftelioglu and S. Shekhar
is Deep Neural Networks. Neural networks that mimic the information flow in a
brain are increasingly getting attention from the computer science research community. Originally proposed in the late twentieth century, research on neural networks stagnated due to the limitation of the hardware at that time. The increased
computational power by GPUs with thousands of cores as well as in memory
matrix manipulations, re-started the neural network research again. Especially in
the computer vision domain, Deep Neural Networks (i.e., deep learning) proved
to be promising.
Some recent applications of deep learning in food–energy–water research
includes identification of objects (trees, farm fields, etc.) from remote sensing imagery, identification of healthy/unhealthy animals from the camera feed in farms as
well as automated detection of ripe fruits from trees. This research is evolving, and
in the future, it is expected to have more use cases for GPU usage for deep learning
purposes. Readers are encouraged to read on these tools but beware that these tools
are often a black box that affects their interpretability.
16.4 Computer Science as a Scientific Tool
The use of computer science in scientific methods is twofold:
Causal and mechanistic models often start with model building and theories. Then
these models are numerically solved and backed by simulations. Many such
models require a complicated process of solving differential equations that are
hard to solve by hand and paper. In addition, often the simulations are hard without the use of a computer. For example, SWAT and GCM techniques require
heavy use of supercomputers.
Data-driven/empirical models involve controlled experiments and statistical analysis. Often, these methods start with hypothesis creation followed by hypothesis
testing as well as hypothesis ranking. Since such approaches are hard to prove by
mathematical theories (due to uncertainties, complications, etc.), they often rely
on a large amount of data in hand (fourth Paradigm of Science). Since these big
datasets require extensive analysis, computational tools that use distributed data
analysis (e.g., big data tools such as Hadoop and Spark) are required to perform
those. For example, the ozone hole over Antarctica was detected by such an
approach.
16.4.1 Nexus Data from Computer Science Perspective
In order to understand the role of computer science in food, energy, and water
research, one may need to understand the different types of data that are related to
nexus research. Since the details of the importance of Data is explained in Chap. 14,
E. Eftelioglu and S. Shekhar