Raju Ram Email & Phone Number

Leonberg, Baden-Württemberg, Germany

Kaiserslautern

In my Ph.D. research, I developed multilevel ILU-based asynchronous and scalable parallel preconditioners in a linear solver library GaspiLS, to solve very large-scale linear systems in simulations. It exploits GASPI based distributed memory parallelism using the Restrictive Additive Schwarz (RAS) method. At the shared memory level, it exploits parallelism with a multilevel nested dissection(MLND) approach. Initial experiments for the convection-diffusion problem show 83% parallel efficiency of RAS preconditioner on 64 processes and 50% parallel efficiency MLND-based approaches, using 64 GASPI processes, each having 20 cores. In Deep learning, I have optimized the Cholesky decomposition in the underlying Hyperparameter Optimization problem using lazy Gaussian Processes. It has improved the run-time by a factor of 15 for MNIST dataset on LeNet5 and 3 for CIFR-10 dataset on RestNet32. I have presented this research at Supercomputing Conference 2019 in Denver, Colorado.

Delft Area, Netherlands

In my master thesis project, I have developed a parallel deflation-based preconditioner to solve large-scale linear systems of equations arising from Groundwater flow equations. It has reduced the run time of the Dutch water management model called Netherlands Hydrological Instrument by 4x on more than 100 MPI processes. The project description can be found at http://ta.twi.tudelft.nl/users/vuik/numanal/ram.html.

Bangalore

I was responsible for software development of the FICO Score:▪ Developed mathematical models to generate FICO and Expansion Score for Banca Transilvania, Romania.▪ Delivered the FICO Score software in C#.NET to our client CRC, Nigeria.▪ Designed a proof of concept for license mechanism software which provides restriction of license usages for the FICO score. For this project, I have developed a variant of Base64 encryption and decryption algorithm to safeguard license key. The programming language used was Java

Eindhoven Area, Netherlands

I was analyzing the PRIMA method which reduces the time complexity of Industrial problems based on Differential Algebraic Equations(DAE). I wrote Matlab code to compute the Index, which measures sensitivity of DAE's. I was investigating why PRIMA fails for DAE's having high index.

Master’S Degree, Applied Mathematics, 8.0/10.0

Master’S Degree, Computational Engineering, 1.5

Bachelor Of Technology (B.Tech.), Mathematics And Computing

Activities and Societies: Swimming, Volleyball

Education record