Permion Platform™
The Permion Platform is our core engine, built through years of development around the notion of employing AI to permit data-oriented computing.
Applications
Permion Arriba
A foundational language for industrial-scale artificial intelligence, Permion Arriba is high-performance, interoperable across languages and systems, and embodies state-of-the-art software engineering best practices and tooling. Permion Arriba combines optimized vector, matrix, and graph processing with logic as its core primitives, enabling support of machine learning and pattern analysis.
Permion Arriba is an environment for bringing many tools to bear on a common problem, allowing the more abstract parts of the application to work in a declarative, logical fashion while the I/O or computation-bound parts of the application can leverage specialized high-performance tools.
Permion Arriba supports graph queries and can translate them into highly efficient operations on sparse matrices. For example, a Mac M1 employing OpenCypher or Neo4J can accomplish a breadth-first traversal of a social network graph of 1.6 billion nodes at a rate of 1.28 billion edges per second. Matrix algorithms can be further accelerated running on GPUs.
Permion Arriba is designed for neuro-symbolic AI programming and can act as an orchestration layer on top of large language models (like GPT). It implements ISO Standard Prolog, supports the Logtalk language for object-oriented and scaled software engineering, comes with an Integrated Development Environment (IDE), and has extensive library support. Permion Arriba interoperates with Python, Java, and other languages, or be invoked using REST or HTTP APIs.
Permion ISA
X-Machines Virtual Machine Instruction Set Architecture (XVM-ISA™ ) is the world’s first fully differentiable graph neuro-symbolic AI compiler, designed to run on diverse platforms from legacy to emerging. XVM-ISA™ combines logical inference with statistical learning in a lightweight package to bring intelligence from the cloud to the edge.
As a virtual machine, the XVM-ISA™ maps Permion programs onto a diverse range of hardware, from low-end and legacy, to contemporary GPU-accelerated cloud hardware, to emerging devices. Permion programs can function on SCADA and internet of things devices all the way up the computing tech stack. This flexibility enables us to achieve fidelity from the cloud to the edge: scientific consistency, application consistency, and stakeholder consistency.
X-machines, also known as input/output automata, are a formal mathematical model used to describe the behavior of concurrent and distributed systems. They consist of a finite set of states, input and output alphabets, and a set of transition functions that define the system's behavior. X-machines are useful for analyzing and verifying the correctness of complex systems, as well as for designing new ones. For Permion, the letter ‘X’ stands for all the data types we support efficiently, as native primitive objects: starting with logic, vectors, matrices, and graphs – creating enormous efficiencies and new possibilities.
As an Instruction Set Architecture, our XVM-ISA™ was designed to allow Permion programs and applications to build and execute on next-generation high-performance computing platforms.
Challenge
Deep learning systems are based on networks of data densely connected, such that the data itself determines how the AI responds to user requests and requires thousands of processors, large amounts of memory, and tremendous power to produce results. Further, data is growing exponentially, with 90% of the world’s data generated in the past 24 months alone.
Permion Solution
Build a core piece of technology—our X-Machines Virtual Machine Instruction Set Architecture (XVM-ISA™)—to bring convergence in security, speed, performance, and power to today’s data requirements.
- World’s first fully differentiable graph neuro-symbolic AI compiler
- Industrial-scale artificial intelligence
- Translate graph queries into efficient sparse matrices operations
- Fidelity from the cloud to the edge
- Designed to run on diverse platforms from legacy to emerging
Featured: Improve Any Graph Neural Network’s Performance
Graph data have complex and irregular structures, making accuracy difficult to achieve and fueling competition to improve machine learning performance. Graph Neural Networks (GNNs) are currently state of the art but still have shortcomings.
Data in graph or network format is one of the most valuable and challenging machine learning (ML) applications in the world today. Applications include social network analysis, recommender systems, discovery of new materials and drugs, as well as emerging tasks in the fields of supply chain, and intelligence analysis. Wrapped around any preexisting GNN-based AI system, FDP reveals hidden relationships and structures in graph data and uses them to improve system performance. We have improved the accuracy of world leading GNNs and are ready to take graph ML to the next level of productivity and impact.