Open-source AI reference kits simplify development

Intel Corp. has released the first set of open-source AI reference kits first introduced at Intel Vision. These kits, designed for on-premises, cloud, and edge environments, include AI model code, end-to-end machine learning pipeline instructions, libraries, and Intel oneAPI components.

(Source: Shutterstock)

These kits are designed to help deploy AI faster and easier in industries such as health care, manufacturing, and retail with higher accuracy, better performance, and lower implementation cost, said Intel.

The AI reference kits are built in collaboration with Accenture. They feature pre-built AI with enterprise contexts for both greenfield AI introduction and existing AI solutions. The four kits available for download include utility asset health, visual quality control, customer chatbot, and intelligent document indexing.

A synopsis of each kit follows below.

• Utility asset health: This predictive analytics model targets higher service reliability for utilities. Key features include an Intel-optimized XGBoost through the Intel oneAPI Data Analytics Library to model the health of utility poles with 34 attributes and more than 10 million data points. Data includes asset age, mechanical properties, geospatial data, inspections, manufacturer, prior repair and maintenance history, and outage records. The predictive asset maintenance model continuously learns as new data, like new pole manufacturer, outages and other changes in condition, are provided, said Intel.
• Visual quality control: Targeting quality control (QC) in manufacturing operations, particularly computer vision applications that require heavy graphics compute power during training and retraining as new products are introduced. The AI Visual QC model was trained using Intel AI Analytics Toolkit, including Intel Optimization for PyTorch and Intel Distribution of OpenVINO toolkit. Both are powered by oneAPI to optimize training and inferencing to be 20% and 55% faster, respectively, compared to stock implementation of the Accenture visual quality control kit without Intel optimizations for computer vision workloads across CPU, GPU and other accelerator-based architectures, said Intel.
• Customer chatbot: Designed for conversational chatbot interactions, the reference kit includes deep learning natural language processing models for intent classification and named-entity recognition using BERT and PyTorch. Intel Extension for PyTorch and Intel Distribution of OpenVINO toolkit optimize the model for better performance. Intel said it provides 45% faster inferencing compared to stock implementation of the Accenture customer chatbot kit without Intel optimizations across heterogeneous architectures. It also allows developers to reuse model development code with minimal code changes for training and inferencing.
• Intelligent document indexing: AI can automate the processing and categorizing of semi-structured and unstructured documents for faster routing and lower manual labor costs. Using a support vector classification (SVC) model, this kit was optimized with Intel Distribution of Modin and Intel Extension for Scikit-learn powered by oneAPI. These tools improve data pre-processing, training, and inferencing times by 46%, 96% and 60% faster, respectively, compared to stock implementation of the Accenture Intelligent document indexing kit without Intel optimizations for reviewing and sorting the documents at 65% accuracy.

These kits can be downloaded for free on the Intel.com AI Reference Kits website. The kits are also available on Github.