Visualization in Scientific Computing

Prof. Dr. Filip Sadlo (Research Group: Visual Computing Group)


In this talk, we investigate existing and potential roles of visualization in scientific computing. Whereas traditional visualization focused on the data resulting from simulation and measurement, current trends aim at a more integral visualization component in scientific computing, ranging from modeling to post-processing. We exemplify these trends with recent projects, and conclude with advances in vector field analysis.

High-order Finite Element Methods in High-performance Computing

Prof. Dr. Peter Bastian (Scientific Computing Group)

The advertised peak performance in floating point operations per second (FLOPS/s) of today's CPUs is very impressive. Even desktop CPUs are capable of more than 10^12 FLOPS/s (1 TFLOP/s). Supercomputers are supposed to reach 10^18 FLOPS/s in the near future. Achieving this performance for real applications is, however, a challenge. It requires 1) that an enormous amount of parallelism is available in the instructions stream and 2) many arithmetic operations are carried out on every datum that is read from main memory. Parallelism needs to be exploited on several levels, in particular on the level of SIMD (vector) instructions.

The talk addresses performance in the context of the numerical solution of partial differential equations using the finite element method. Matrix-free implementation of high-order methods offers the possibility of bypassing the memory bottleneck while at the same time reducing the number of floating-point operations substantially. Numerical results will demonstrate that a substantial fraction of peak performance can be achieved for operator application. In  addition preconditioning as well as parallel scalability to thousands of cores will also be addressed.

Identification of Humpback Whales using Deep Metric Learning

Artsiom Sanakoyeu (Heidelberg Collaboratory for Image Processing)


Learning the embedding space, where semantically similar objects are located close together and dissimilar objects far apart, is a cornerstone of many computer vision applications. 

I will discuss existing approaches to learn the embedding space using Deep Metric Learning (DML) as long as our novel `divide and conquer` approach for deep metric learning, which significantly improves the state-of-the-art performance of metric learning. 

In the second part of the talk, I will show how deep metric learning approaches can be applied to a real-world problem: Humpback whale identification. 
To aid whale conservation efforts, scientists use photo surveillance systems to monitor ocean activity. They use the shape of whales' tails and unique markings found in footage to identify what species of whale they’re analyzing and meticulously log whale pod dynamics and movements.
Humpback whale identification challenge was hosted at Kaggle platform. More than 2100 teams were challenged to build a computer vision algorithm to identify individual whales in images and release biologists form tedious manual work. I will discuss our solution based on DML which placed us in the Top-10 in the final standings.


SIAM Colloquium Summer Term 2019

In the summer term 2019 we host the SIAM Colloquium for the third time. The schedule is:


Prof. Dr. sc. Filip Sadlo

Visualization in Scientific Computing

7th May

Prof. P. Bastian

High-order Finite Element Methods in High-performance Computing

4th June

Artsiom Sanakoyeu

Identification of Humpback Whales using Deep Metric Learning

9th July