Big data is much more than storage of and access to data. Analytics
plays an important role in making sense of that data and exploiting its
value. But learning from big data has become a significant challenge and
requires development of new types of algorithms. Most machine learning
algorithms encounter theoretical challenges in scaling up to big data.
Plus there are challenges of high dimensionality, velocity and variety
for all types of machine learning algorithms. The neural network field
has historically focused on algorithms that learn in an online,
incremental mode without requiring in-memory access to huge amounts of
data. The brain is arguably the best and most elegant big data processor
and is the inspiration for neural network learning methods. Neural
network type of learning is not only ideal for streaming data (as in the
Industrial Internet or the Internet of Things), but could also be used
for stored big data. For stored big data, neural network algorithms can
learn from all of the data instead of from samples of the data. And the
same is true for streaming data where not all of the data is actually
stored. In general, online, incremental learning algorithms are less
vulnerable to size of the data. Neural network algorithms, in
particular, can take advantage of massively parallel (brain-like)
computations, which use very simple processors, that other machine
learning technologies cannot. Specialized neuromorphic hardware,
originally meant for large-scale brain simulations, is becoming
available to implement these algorithms in a massively parallel fashion.
Neural network algorithms, therefore, can deliver very fast and
efficient real-time learning through the use of hardware and this could
be particularly useful for streaming data in the Industrial Internet.
Neural network technologies thus can become significant components of
big data analytics platforms and this special issue will begin that
journey with big data.
For this special issue of Neural Networks,
we invite papers that address many of the challenges of learning from
big data. In particular, we are interested in papers on efficient and
innovative algorithmic approaches to analyzing big data (e.g. deep
networks, nature-inspired and brain-inspired algorithms),
implementations on different computing platforms (e.g. neuromorphic,
GPUs, clouds, clusters) and applications of online learning to solve
real-world big data problems (e.g. health care, transportation, and
electric power and energy management).
RECOMMENDED TOPICS:
Topics of interest include, but are not limited to:
- Autonomous, online, incremental learning – theory, algorithms and applications in big data
- High dimensional data, feature selection, feature transformation – theory, algorithms and applications for big data
- Scalable neural network algorithms for big data
- Neural network learning algorithms for high-velocity streaming data
- Deep neural network learning
- Neuromorphic hardware for scalable neural network learning
- Big data analytics using neural networks in healthcare/medical applications
- Big data analytics using neural networks in electric power and energy systems
- Big data analytics using neural networks in large sensor networks
- Big data and neural network learning in computational biology and bioinformatics
SUBMISSION PROCEDURE:
Prospective authors should visit
http://ees.elsevier.com/neunet/
for information on paper submission. During the submission process,
there will be steps to designate the submission to this special issue.
However, please indicate on the first page of the manuscript that the
manuscript is intended for the Special Issue: Neural Network Learning in
Big Data. Manuscripts will be peer reviewed according to Neural
Networks guidelines.
Manuscript submission due: December 15, 2014
First review completed: March 1, 2015
Revised manuscript due: April 1, 2015
Second review completed, final decisions to authors: April 15, 2015
Final manuscript due: April 30, 2015
GUEST EDITORS:
