OASIS: Open-source AI Software Infrastructure for Science

Recent improvements in artificial intelligence (AI) have made a huge impact on consumer product development, usage, and experiences as seen in products such as Amazon Alexa and IROBOT. While AI has been very successful in the consumer market, its growth has lagged in some science domains. Many reasons exist for the delayed adoption of AI in the science domain and from the software perspective, the key one is the availability of ready-to-use AI software infrastructure for sciences. Current AI tools are hardcoded to industrial data types and there- fore a great deal of effort is required to make scientific data AI-ready. Also, the availability of large scientific datasets and AI’s need for large training datasets mean training of models needs to run at scale. This requires domain experts to interface with a complex software stack that comprises many computer infrastructure pieces such as containerization, one or more machine learning frameworks, and I/O packages. We propose to build an open source software platform, OASIS: Open-source AI Software Infrastructure for Science that is designed for researchers in the scientific domain. The innovative features of OASIS include a data API to directly ingest and serve scientific data formats and annotations in the AI workflow, a unique integration of features such as coupling of data to AI models, scalable training, HPC or cloud deployment into a cohesive web and command-line interface, and standard and state-of-the-art techniques to debug and enhance AI models. OASIS will make it easy to consume 1D to 4D datasets by developing new data interfaces that will bridge AI tools to scientific data formats. It will provide data pipelines and data processing to make efficient data representation of scientific datasets. Finally, OASIS will provide an integrated web and command line interface to upload raw data and annotations, serialize and upload models, run training and inference at scale, and visualization tools to help understand model outcome and performance. This work will further our open source platforms for data integration, AI, and visualization while providing a turn-key platform for scientific com- munities. Such platforms provide Kitware with significant business opportunities, as open soft- ware platforms lend themselves to a services business model where the company can provide technology integration services with customers and collaborators to build valuable, competitive products and services.