Intel releases open source AI reference kit

Open-source design simplifies AI development for solutions in healthcare, manufacturing, retail, and other industries

Intel has officially launched its first open source AI reference kit, designed to make it easier for enterprises to deploy AI locally, in the cloud, and in edge environments. These reference kits, first announced at the Intel Vision Summit, include AI model code, end-to-end machine learning pipeline instructions, libraries, and Intel oneAPI components for cross-architecture operation, allowing data scientists and developers to learn how to deploy AI with higher accuracy, better performance, and lower total cost of implementation in healthcare, manufacturing, retail, and other industries more quickly and simply.

Dr. Wei Li, vice president and general manager of Intel’s Artificial Intelligence and Analytics Division, said: “Innovation can only flourish in an open and crowd-innovative environment. Whether it is the Intel Accelerated Open AI Software Ecosystem, which includes a variety of optimized popular frameworks, or Intel’s AI tools, they are all built on the open, standards-based, and unified oneAPI programming model. The reference kit launched today is built with Intel’s end-to-end AI software products, which will enable millions of developers and data scientists to easily and quickly add AI to applications or improve existing intelligent solutions.”

AI workloads are growing and becoming more diverse as use cases emerge in areas such as vision, speech, and recommendation systems. The Intel AI Reference Kits, developed in conjunction with Accenture, are designed to accelerate the adoption of AI across industries. These kits are open source, pre-built AIs that support the introduction of new AIs and strategic adjustments to existing AI solutions for a variety of important enterprise application scenarios.

Intel will release four kits for download:

● Utility asset health: As global energy consumption continues to grow, the number of power transmission assets is expected to grow as well. This predictive analytics model was trained to improve the service reliability of utilities. Using the Intel® oneAPI Data Analytics Library, it models the condition of utility poles based on 34 attributes and over 10 million data points using the Intel-optimized XGBoost algorithm1. Data types include asset age, mechanical properties, geospatial data, inspection reports, manufacturers, previous repair and maintenance history, and outage records. The predictive asset maintenance model continuously learns from newly provided data, such as new pole manufacturers, outages, and other changes in conditions.

● Visual Quality Control: Quality control is an essential part of all manufacturing operations. The challenge with computer vision techniques is that they often require a lot of image processing power during training and need to be retrained frequently as new products are introduced. This AI visual quality control model was trained using the Intel® AI Analytics Toolkit, including Intel® PyTorch optimization, and the Intel® Distribution of OpenVINO™ toolkit, both powered by oneAPI. For computer vision workloads across CPUs, GPUs, and other accelerator-based architectures, this model is 20% faster in training and 55% faster in inference than the existing Accenture Visual Quality Control Suite that is not Intel-optimized2. Using computer vision techniques and the SqueezeNet classification algorithm, this AI visual quality control model can detect drug defects with 95% accuracy through hyperparameter tuning and optimization.

● Customer Service Robot: Conversational chatbots have become a critical service to support the development of the entire enterprise. The AI ​​models used for conversational chatbot interactions are large-scale and highly complex. This reference kit includes deep learning natural language processing models for intent classification and named-entity recognition, using BERT and PyTorch. The Intel® Extension for PyTorch and Intel® Distribution of OpenVINO™ tools optimize the model for higher performance across heterogeneity, with a 45% increase in inference speed compared to the existing Accenture Customer Service Robot Kit that has not been optimized by Intel3, while allowing developers to reuse model development code for training and inference with minimal code changes.

● Intelligent Document Indexing: Enterprises need to process and analyze millions of documents every year, and many semi-structured and unstructured documents require manual operations. AI can automatically process and classify these documents to increase speed and reduce labor costs. This kit uses a support vector classification (SVC) model and is optimized through the Intel® distribution Modin and Intel® Extension for Scikit-learn supported by oneAPI technology. Compared with the existing Accenture Intelligent Document Indexing Toolkit that has not been optimized by Intel4, these tools will increase data preprocessing, training, and inference time by 46%, 96%, and 60%, respectively, and can review and analyze documents with 65% accuracy.

These AI reference kits can be downloaded for free on the AI ​​reference kit page on Intel's official website or on Github.

Developers want to add AI to their solutions, and the AI ​​reference kits released by Intel this time will help achieve this goal. These kits are built on Intel's end-to-end tools and frameworks to optimize AI software and complete this product portfolio. Based on oneAPI's open, standards-based, heterogeneous programming model development that can run on multiple architectures, these tools can overcome the limitations of proprietary environments and help data scientists train models faster and at a lower cost.

In the coming year, Intel will also release a series of new open source AI reference kits, providing a variety of trained machine learning and deep learning models to help companies of all sizes achieve digital transformation.