Source author record

Detlef Hohl

Detlef Hohl appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

cond-mat.mtrl-sci cond-mat.soft Distributed, Parallel, and Cluster Computing Machine Learning

Catalog footprint

What is connected

3works

4topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2020arXiv

Deep Neural Network Learning with Second-Order Optimizers -- a Practical Study with a Stochastic Quasi-Gauss-Newton Method

Training in supervised deep learning is computationally demanding, and the convergence behavior is usually not fully understood. We introduce and study a second-order stochastic quasi-Gauss-Newton (SQGN) optimization method that combines ideas from stochastic quasi-Newton methods, Gauss-Newton methods, and variance reduction to address this problem. SQGN provides excellent accuracy without the need for experimenting with many hyper-parameter configurations, which is often computationally prohibitive given the number of combinations and the cost of each training process. We discuss the implementation of SQGN with TensorFlow, and we compare its convergence and computational performance to selected first-order methods using the MNIST benchmark and a large-scale seismic tomography application from Earth science.

preprint2016arXiv

Performance Analysis and Optimization of a Hybrid Distributed Reverse Time Migration Application

Applications to process seismic data employ scalable parallel systems to produce timely results. To fully exploit emerging processor architectures, application will need to employ threaded parallelism within a node and message passing across nodes. Today, MPI+OpenMP is the preferred programming model for this task. However, tuning hybrid programs for clusters is difficult. Performance tools can help users identify bottlenecks and uncover opportunities for improvement. This poster describes our experiences of applying Rice University's HPCToolkit and hardware performance counters to gain insight into an MPI+OpenMP code that performs Reverse Time Migration (RTM) on a cluster of multicore processors. The tools provided us with insights into the effectiveness of the domain decomposition strategy, the use of threaded parallelism, and functional unit utilization in individual cores. By applying insights obtained from the tools, we were able to improve the performance of the RTM code by roughly 30 percent.

preprint2009arXiv

Theoretical investigation of methane under pressure

We present computer simulations of liquid and solid phases of condensed methane at pressures below 25 GPa, between 150 and 300 K, where no appreciable molecular dissociation occurs. We used molecular dynamics (MD) and metadynamics techniques, and empirical potentials in the rigid molecule approximation, whose validity was confirmed a posteriori by carrying out it ab initio MD simulations for selected pressure and temperature conditions. Our results for the melting line are in satisfactory agreement with existing measurements. We find that the fcc crystal transforms into a hcp structure with 4 molecules per unit cell (B phase) at about 10 GPa and 150 K, and that the B phase transforms into a monoclinic high pressure phase above 20 GPa. Our results for solid/solid phase transitions are consistent with those of Raman studies but the phase boundaries estimated in our calculations are at higher pressure than those inferred from spectroscopic data.