2019 ESIP Winter Meeting

This panel will focus on scalable architectures for implementing machine learning and AI models across Earth Sciences focus areas. The specific details would include both HPC and cloud-centric operational pipelines for automated data processing, parallel multi-GPU model implementations, cluster scheduling, resource optimization and time to delivery - a few tools of discussion can focus around DC/OS, Terraform/Nomad, and Kubernetes. In addition to compute constraints, the panel also encourages discussion on ML framework optimizations and operational complexity, including available frameworks and workflow tools like Tensorflow, Mxnet, Pytoch, Google ML pipeline, AWS Sagemaker and others. GPU hardware optimization is another topic of interest where the panel will focus on how new generation GPU’s (with more CUDA cores) are able to perform faster training as opposed to old generation GPUs.