the scope of this article, but the reader is encouraged to
explore prior publication for more detailed explanations
of our multidimensional analysis techniques [113,120].
EDEN is an exemplary case of the indispensable visual analytics
techniques that provide intelligent user interfaces
by leveraging both visual representations and human
interaction, thereby enhancing scientific discovery with
vital assistance from automated analytics. As we develop
new visual analytics approaches like EDEN for materials
science workflows, we expect to dramatically reduce
knowledge discovery timelines through more intuitive and
exploratory analysis guided by machine learning algorithms
in an intelligent visual interface.
Conclusions
The development of electron and scanning probe microscopies
in the second half of the twentieth century
was enabled by computer-assisted methods for automatic
data acquisition, storage, analysis, and tuning and refinement
of feedback loops as well as imaging parameters. In
the last decade, high-resolution STEM and STM imaging
techniques have enabled acquisition of high-veracity information
[121] at the atomic scale, readily providing insight
on positions and functionality of materials that have been
inaccessible due to a lagging analysis framework in the microscopy
communities. Naturally, progress in complexity
of dynamic and functional imaging leads to multidimensional
data sets containing spectral information on local
physical and chemical functionalities, which can be easily
expanded further to acquire data as a function of a plethora
of parameters such as time, temperature, or many
other external stimuli.
Maximizing the scientific output from existing and future
microscopes brings forth the challenge of analysis,
visualization, and storage of data, as well as decorrelation
and classification of the known and unknown hidden data
parameters, the traditional big data analysis. The existing
infrastructure for such analysis has been developed in the
context of medical and satellite imaging, and its extension
to functional and structural imaging data is a natural next
step. Of course, further development toward a flexible infrastructure
where the scientists can select or define their
own analysis algorithms to analyze the data ‘on the fly’ as
it is being collected can be envisioned. This will require
scalable algorithms, high-performance computing, and
storage infrastructure. Reducing the data sets to a more
manageable size, while initially attractive, comes with the
risk of losing significant information within the data, particularly
for exploratory studies in which the phenomena
of interest may not be captured by statistical methods.
Beyond the big data challenges [122,123] is the transition
to a deep data approach, in which we fully utilize all
the information present within the data to derive an understanding
[124] - namely, how do we ascribe relevant