Increasingly large data sets are being ingested and produced by simulations. What experience from large-scale simulation is transferable to big data applications? Conversely, what new optimal algorithms will emerge that are motivated by data-intensive applications being pushed to large scales? How will they enrich traditional simulation? As long as the software stacks, production facilities, and even developer and user communities remain separate, many opportunities for mutual enhancement will be unrealized. This workshop will discuss:
benefits of in situ convergence of simulation, analytics, and machine learning
steering in high-dimensional parameter space
smart data compression
data-driven modeling (e.g., refinement of empirical functions through learning)
physics-based “regularization” of analytics
simulation as a source of training data
learning to impute missing data
evolving requirements of high-performance analytics and simulation
scalable hierarchical algorithms for analytics and simulation
detecting and exploiting data sparsity
open problems, where no scalable methods yet exist
The workshop will bring together analysts and developers of computationally and data-intensive applications interested in early exploitation of extreme-scale computing platforms to define common ground and seek new opportunities.