Scalable Data Analytics in Scientific Computing (SDASC) workshop invites submissions of original research. More details available at: SDASC web site.

The ever increasing importance of methods originating in statisticalinference and their growing use at large cloud computing facilitiesleads both scientific and HPC communities to look into new ways ofapplying computational steering and incorporate it into theirlarge-scale simulations. The SDASC workshop will feature automated dataanalysis efforts at the convergence of computational science, HPC,large-scale data analytics and inference. The focus will be on theintegration of the HPC techniques and statistical learning tasks intothe modern software stack of modern computational science.

The half-day SDASC workshop will gather experts from the computational science, HPC, and machine learning communities. The committee members are recognized in their respective fields as experts of note and will assure fulfilment of the goals of the workshop.

Submission Guidelines

All papers must be original and cannot be simultaneously submitted toanother journal or conference or be in review process for another event.

The workshop will use single-blind peer review.  The submitted manuscripts will be reviewed anonymously but the authors will be known to the reviewers. Submissions will be scored on the following criteria: originality, technical strength and correctness as well as significance, quality of presentation, and relevance to the workshop topics of interest.

List of topics

• Scientific data set creation, ingest, curation, labelling, and  analysis with statistical models and inference

• Incorporating realtime and ad-hoc data analytics into applications and  their deployment on supercomputing and cluster platforms

• Computational steering through machine learning models and related  control  theory approaches

• Meta-data and data metrics collection and generation for large data  collections and output data sets of computational simulations

• Multi-precision training/inference methods and their use on modern  hardware for simulation data

• Novel use of discriminative and generative machine learning   approaches for scientific data sets including Adversarial and   Reinforcement Learning with self-supervision

• Modern HPC storage issues when dealing with integration of  computational simulation outputs with data analytics software

• Synchronous and asynchronous learning approaches at scale for methods  related to deep neural network training, stochastic gradient descent,  loss-function engineering, and related distributed optimization  techniques

• Model derivation and training for scalable simulations and data sets

• Hyperparameter search and optimization incorporating recent advances  in Bayesian optimization

• Deployment of statistical models and their implementations such as  TensorFlow and PyTorch or application-specific tensor frameworks.

• Integration of models with large scale simulations code bases through  containers (Kubernetes, Docker, Singularity, OpenShift),  virtualizaiton, colocation, and workflow frameworks

We also welcome cross-cutting submissions that are span some of the topics mentioned above.

Committees

Program Committee

• Gabriele Cavallaro (Juelich Supercomputing Centre, Germany)
• Marat Dukhan (Google Inc., USA)
• Eileen Kūhn (Karlsruhe Institute of Technology, Germany)
• Daniel Jacobson (Oak Ridge National Laboratory, USA)
• Xipeng Shen (North Carolina State University, USA)
• Martin Siggel (German Aerospace Center /DLR/ Cologne, Germany)
• Misha Smelyanskiy (Facebook Inc., USA)
• Miroslav Stoyanov (Oak Ridge National Laboratory, USA)

Organizing committee

• Piotr Luszczek, University of Tennessee, USA
• Hartwig Anzt, Karlsruhe Institute of Technology, Germany

Local Organizing committee

• Markus Götz (Karlsruhe Institute of Technology, Germany)

Submission Deadlines

• Submission deadline: April 29, 2021 (AoE)

Publication

The accepted papers will be published in Springer LNCS proceedings.

Manuscripts should be 12 pages maximum excluding the references (we encourage authors to include relevant references). Papers need to be formatted according to Springer's single column LNCS style (see LaTeX and Word templates).

Note: 12 pages LNCS is roughly equivalent to 6 pages in double column IEEE format.

Venue

The workshop is co-located with ISC High Performance 2021 and will be be hold virtually. No in-person meeting is planned.

Contact

All questions about submissions should be emailed to the address listed on the EasyChair submission site.