A complete two day face-to-face meeting in Iceland in the DEEP-EST project. Co-designing the next generation supercomputer prototype is no easy task, especially given the vastly different applications from the software application partners, such as the University of Iceland which I represent on the project. The meeting was a splendid success and I can´t wait to run my applications on all of the new hardware (which I can´t tell you about yet)
I presented a poster at EASC18, which included a lightning-talk, detailing our plans on enhancing Deep Learning (DL) with a modular super-computing approach which is being developed in the DEEP-EST project.
Here´s a PDF version of my poster if anyone is interested.
EASC2018_Poster_A0
On a side not, Edinburgh is a beautiful city, and I quite enjoyed the haggis :).
Had an oral presentation on my conference abstract paper at the EGU18 on “Scaling DBSCAN towards exascale computing for clustering of big data sets” where I presented the challenges we face in the DEEP-EST project with clustering truly massive point-cloud data sets in the Terabyte range.
A do-over of the three day, hands-on, machine learning workshop in the Jülich Supercomputing Centre (JSC), in Germany, which I co-supervised. The popularity of the last course from January was so high that it was decided that we should host it again, filling the 40 attendee slots immediately. Attendees learned about the basics of machine learning, and extensively about SVMs, performing training, testing, cross-validation, and grid-search, on the Jureca Supercomputer.
A three day, hands-on, machine learning workshop in the Jülich Supercomputing Centre (JSC), in Germany, which I co-supervised. Attendees learned about the basics of machine learning, and extensively about SVMs, performing training, testing, cross-validation, and grid-search, on the Jureca Supercomputer.
A three day workshop hosted by the Barcelona Supercomputing Center (BSC) where I gained some hands-on knowledge on their performance tools, e.g. Dimemas and Paraver, which are used to profile High-Performance-Computing (HPC) applications.
Sadly, I was not able to visit the MareNostrum supercomputer, which is located in an old cathedral in Barcelona (picture above).
My talk with the title “Building a P2P Darknet App: Lessons learned” was accepted at the 2017 Droidcon conference in Berlin, one of the largest independent Android developer event in the world.
In my talk I outlined some of the challenges we faced during the development of the Briar project, an open-source secure communication framework that uses darknet to hide all the metadata.
My Talk is available in its entirety on YouTube
Today marks the beginning of DEEP-EST, a three year High-Performance-Computing (HPC) research project which is funded by the European Commission with the Horizon 2020 research initiative. I will be directly involved in this project as a representative of the University of Iceland, which is one of six application partners. My task will be to map ,and optimize, three machine learning applications; Support Vector Machines, Highly Parallel DBSCAN, and a custom Deep Learning application; to a heterogeneous supercomputer prototype.
The Image above was taken during the kick-off meeting at the Jülich Supercomputing Centre in North Rhine-Westphalia, in Germany, which coordinates the project.
More information on DEEP-EST, and the DEEP projects can be found here and here.
Ernir Erlingsson 