This post introduces the Acumos AI platform, an open source framework to simplify the development of artificial intelligence and machine learning applications.
Acumos wraps tools, such as TensorFlow and SciKit Learn, and models with a common API that allows them to seamlessly connect. It is language agnostic, supporting tools and models built in a variety of popular software languages. Acumos also is designed to leverage modern microservices. Anyone building an AI application on Acumos can export the libraries, models and all other required information as Docker files. Lastly, Acumos includes the Acumos Marketplace – a shopping portal from which to browse a catalog of AI models contributed by the community. To make Acumos incredibly easy to use, the platform also contains a visual editor with drag-and-drop functionality.
Overall, we hope that Acumos will enable a massive acceleration in AI adoption and innovation. Please read on!
# # #
By Reuben Klein, AT&T
Over the last several years, there has been tremendous excitement around and interest in the potential for machine learning (ML) technologies applied to artificial intelligence (AI). Tech giants and venture capitalists have invested tens of billions of dollars in building AI technologies. But we are still in the very early stages of AI development. Unfortunately, the talent to build AI remains scarce and expensive. An analysis by Tencent found that there are only 300,000 engineers in the world with the skills necessary for AI development. Compiling and cleaning the massive amounts of data required to make AI applications work is also expensive and time consuming. Constructing and maintaining all the infrastructure required for AI is also costly and time consuming.
In reality, most of the organizations building AI applications are using the same software and tools to create custom AI environments. This is because the ecosystem lacks common, easy-to-use frameworks for developing AI applications. Contributors to the Acumos AI Project have felt this same pain. This same problem – lack of talent, expensive to build – prevents many organizations from exploring AI applications even though they have strong data science or analytics capabilities.
To address this gap and make AI accessible to a much wider group of people, including within our own organizations, we built the Acumos AI platform. An open source platform, Acumos simplifies and streamlines the training, integration, and deployment of AI models. Acumos provides a visual editor that allows AI microservices to be chained together to form a robust application with automatic API matching from each component to the next.
By creating an open source community around deep learning, artificial intelligence, and machine learning, The Linux Foundation hopes to accelerate the transition to AI-based software across a wide range of industrial and commercial problems. Uses cases we aim to address with Acumos span telecommunications and networking, advertising and media, academic research, and the Internet of Things, to name a few. Because it is so easy to onboard new tools and models into Acumos, there are a nearly infinite number of potential use cases.
A Platform for Builders, Creators, and Users
At its core, Acumos is designed to simplify and accelerate AI for three key groups. For builders of AI infrastructure, Acumos is designed to streamline setup and reduce the need to maintain complex build and deploy systems. For creators of AI models – data scientists and others – Acumos empowers them to publish models while shielding them from the need to custom develop fully integrated AI solutions. Acumos creates an extensible mechanism for packaging, sharing, licensing, and deploying AI models and publishes them in a secure catalog that is easily distributed between peer systems. It packages each model into an independent, containerized microservice, which is fully interoperable with any other Acumos microservice, regardless of whether it was built with TensorFlow, SciKit Learn, RCloud, H2O, or any other supported toolkit. Models built with any of these tools or any supported language, including Java, Python, and R, can be automatically onboarded, packaged, and cataloged.
For users of AI – the people and organizations using AI to add machine intelligence into their applications – Acumos microservices are easy to integrate into practical applications. A visual editor allows any software developer, even without a background in data science or knowledge of any specialized AI development tools, to construct simple or chained AI applications in Acumos and deploy them as containers to any suitable external environment that supports Docker.
A Community-Driven Marketplace for Models and Solutions
To tap the power of the AI community, we included the Acumos Marketplace, where members can share models and test data in a secure environment. The Acumos Marketplace contains mechanisms for data-powered decision making and artificial intelligence solutions. Self-organized peer groups across one company or across multiple domains can securely share information among themselves on how AI solutions perform, with ratings, popularity statistics, and user-provided reviews. The Acumos platform and Acumos Marketplace will aid communication between data scientists and application developers in order to automate the process of user feedback for selecting models while automating error reporting and software updates for models that have been acquired and deployed through the marketplace.
A Federated Model for AI Development and Deployment
After great deliberation, we decided to use a federated model for Acumos. Here’s how it works. Organizations wishing to deploy the Acumos platform can download the software and run it on the infrastructure of their choice. They can run it as a standalone node or they can join multiple Acumos nodes together. They can even choose to join Acumos nodes with other organizations or partners. For models and solutions published in their local Acumos Marketplace catalogs, they can elect to share them globally with all Acumos users, with federated groups of Acumos users or only with Acumos users on that node. The sharing granularity and security will allow groups to share data to solve problems more effectively. For example, a group of telecommunications and security companies may share network DDOS signature data or models used to detect DDOS to benefit the entire community.
Making AI Accessible to All
With Acumos, software developers can transition from code-writing and editing AI toolkits into a classroom-like code training process. AI models can be acquired from the marketplace, trained, graded on their ability to analyze datasets, and integrated, automatically, into completed solutions. They can access encapsulated AI models, without knowing the details of how they work, and connect them to a variety of data sources, using a range of data adaptation brokers, to build complex applications through a simple chaining process.
Acumos also includes a visual design tool called Design Studio for chaining together multiple models, data translation tools, filters, and output adapters into full end-to-end solutions that can be packaged and deployed into any run-time environment, such as a cloud service, including AIC, Azure, and AWS, a private datacenter, or a specialized hardware environment designed to accelerate AI applications. Acumos only requires a container management facility, like Docker, to deploy and execute portable general-purpose AI applications.
We hope that you will download Acumos, contribute models and tools, and join our community. Your participation will make AI accessible to all.