The world has been changed by the usage of Artificial Intelligence (AI) to quickly understand data as it relates to our environments. These AI workloads are enabled through the use of Machine Learning (ML) and Deep Learning (DL) frameworks allowing insight into how information can be learned by computers to find solutions and answer questions. Many times, these ML/DL frameworks are very difficult to install and manage disabling the technologies like GPGPU many companies and researchers need to overcome limits. Computational researchers are constantly fighting the need to use new tools and enable them versus using the tools to answer scientific questions. Most researchers would rather spend time answering questions over having to install or update software packages that enable the hardware they use. We have found if research groups are enabled to manage these tools with little impact to their time or effort, they are more likely to use them
Recently Open-CE was established as a new community driven set of ML/DL tools enabling some of the best hardware with low activation energy for end users. Open-CE was built off of Watson Machine Learning Community Edition (WML-CE) provided by IBM and OpenPOWER. This new version has a similar set of tools but can now be controlled and managed by the community that uses and needs the resource. Because there will still be a separation on developers who will want full control and end users who want just compiled binaries the Open-CE community is working to provide both. The main Open-CE Github page (https://github.com/open-ce) will be focused on providing feedstock for developers and groups like OSUOSL and the CGRB at Oregon State will provide pre-compiled Conda packages (https://osuosl.org/services/powerdev/opence).
Open-CE is valuable to researchers because it provides the latest and greatest AI package and framework versions pre-integrated in an easy to consume and use conda environment. Cutting edge research requires cutting edge tools and has recently been proven by a recent publication around passive monitoring of animals in the forest entitled "Workflow and convolutional neural network for automated identification of animal sounds" (https://doi.org/10.1016/j.ecolind.2021.107419). Lesmeister lab in the USFS along with the Levi Lab and CGRB at Oregon State have found using the OpenPOWER hardware for both segmentation and classification of data increased the throughput allowing them to change the scope of work. Initially this work was focused on passive monitoring of the norther spotted owl population, however with increased throughput we were now able to start looking through the same data to monitor more species. Table 1 below from the paper shows the ability to monitor different species. The important part of this forward progress within this research is the ability of the computational scientist processing the data having the control over the tools including the optimized ML/DL work flows needed to accomplish this work.
Open-CE has enabled this capability, reduced management of technologies, while providing cutting edge tools for research like this. We plan to start looking at how to bring this processing closer to the edge where the data is collected and having Open-CE be cross platform will help maintain continuity.