Karl Nyberg's Current Research Projects
Karl Nyberg's Current Research Projects
1) Data Science Bowl 2017, run by
Kaggle.
a) Massive data sets. They are DICOM images of lungs, looking for cancer.
b) Particularly interesting was how one competitor was able to use analysis of submission result scores
to determine the answer set for stage one WITHOUT DOING ANY ACUTAL ANALYSIS OF THE IMAGES.
2) Playing with RF again -
MITRE Internet of Things They didn't get a lot of submissions
but one of them was good enough to win the $50,000 prize
a) It was interesting and a bit educational.
3) Converting ASTER image data into DEM data.
Yes, I know
they've already done it -
ASTER GDEM, but I'm doing it for a learning exercise, and to generate
DEM data that I can use for other applications that theirs can't be used
for.
a) Automating the generation of a DTD specific XML parser. (done,
paper
publication anticipated August 2010)
b) Creating a native HDF parser in Ada. (subset implementation done,
paper
publication anticipated August 2010)
c) Writing stereo-correlation software for doing the DEM generation (code in
process, paper in process)
a) Writing code and a paper on a parallel implementation of the
Karatsuba algorithm.
b) Writing a paper on the appropriate application of multicore technology to
efficiently utilize the underlying computational capabilities. (Think micro- v
macro-computing: "automatic" parallelizing compilers only get localized
use of multiple cores; architectural / design modifications can get you
globalized use.)
c) Writing a paper on the overhead associated with multicore computing and
how it constrains / limits the effective throughput of the hardware.
Proposal work in process, funding sources sought (offers welcome! :-)).
Some companies that produce multicore / HPC equipment willing to let me test
on their platforms!
a) migrating from decimal arithmetic (base 10) to byte arithmetic (base 2^8
- 256) done. Known existing solutions confirmed in test configuration.
Performance testing and analysis.
b) migrating from decimal arithmetic (base 10) to byte arithmetic (base 2^16
- 65536)
Parallelizing above to convert serial implementations to multicore / network
configurations.