Adviser: An Intuitive Multi-Cloud Platform for Scientific and ML Workflows

A team of researchers including Shihan Cheng, Michael A. Laurenzano, and David A. B. Hyde has published a paper introducing Adviser, a new platform designed to democratize access to high-performance cloud computing for scientific and machine learning applications. The core problem Adviser tackles is the significant expertise gap required to leverage modern cloud resources, which spans configuring complex software, navigating thousands of instance types and pricing models, and managing distributed execution. Their study confirms this gap remains a major barrier to computational science.

Adviser's solution is an intuitive platform centered on a workflow abstraction. These workflows are reusable artifacts crafted by domain experts that encapsulate all necessary steps—from environment setup and data processing to simulation, result capture, and visualization. Users simply specify their high-level scientific intent, and Adviser takes over the heavy lifting: it automatically provisions the optimal cloud resources, handles runtime configuration, and manages data movement across multiple cloud providers. The platform was validated using two computational glaciology codes, Icepack and PISM, demonstrating how scientists can gain insights and perform rapid exploration of cost-performance trade-offs and scaling behavior without needing deep expertise in cloud infrastructure or high-performance computing.