Intel releases first set of open-source AI benchmark kits


Intel says the first set of open-source AI reference kits are designed to make AI more accessible to organizations in on-premises, cloud, and edge environments.

First introduced at Intel Vision, the reference kits include AI model code, end-to-end machine learning pipeline instructions, libraries, and Intel oneAPI components for cross-architecture performance.

Intel says these kits allow data scientists and developers to learn how to deploy AI faster and easier in healthcare, manufacturing, retail and other industries with greater precision. better performance and lower total cost of implementation.

Wei Li, Intel vice president and general manager of AI and analytics, says innovation thrives in an open and democratized environment.

“Intel’s accelerated open AI software ecosystem, including popular optimized frameworks and Intel AI tools, is built on an open, standards-based, and unified oneAPI programming model. he says.

“These reference kits, built with components from Intel’s end-to-end AI software portfolio, will enable millions of developers and data scientists to quickly and easily introduce AI into their applications or strengthen their existing smart solutions.

Intel’s AI Reference Kits, co-designed with Accenture, are designed to accelerate AI adoption across industries. The four kits available for download include:

1. Utility Asset Health

As power consumption continues to grow globally, field power distribution assets are expected to grow. Intel says this predictive analytics model helps utilities deliver greater service reliability. It uses XGBoost powered by Intel through the Intel oneAPI data analysis library to model utility pole health with 34 attributes and over 10 million data points.

Data includes asset age, mechanical properties, geospatial data, inspections, manufacturer, previous repair and maintenance history, and failure records. The predictive asset maintenance model continuously learns as new data, such as new pole manufacturer, outages, and other status changes, is provided.

2. Visual quality control

Quality control (QC) is essential in any manufacturing operation. Intel says the challenge with computer vision techniques is that they often require significant graphics computing power during training and frequent 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 powered by oneAPI.

Intel says it’s about optimizing training and inference to be 20% and 55% faster, respectively, compared to the stock implementation of Intel’s visual quality control kit. Accenture without Intel optimizations for computer vision workloads on CPUs, GPUs, and other accelerator-based architectures. Using computer vision and SqueezeNet classification, the AI ​​Visual QC model used hyperparameter tuning and optimization to detect defects in pharmaceutical pills with 95% accuracy.

3. Chatbot customer

Conversational chatbots have become an essential service to support initiatives across the enterprise. Unfortunately, Intel claims that the AI ​​models that support conversational interactions with chatbots are very complex. This reference kit includes deep learning natural language processing models for intent classification and named entity recognition using BERT and PyTorch.

Intel says its Intel extension for PyTorch and the Intel Distribution of OpenVINO toolkit have optimized the model for better performance. For example, it has 45% faster inference compared to the stock implementation of Accenture’s Customer Chatbot Kit without Intel optimizations.

4. Intelligent Document Indexing

Companies process and analyze millions of documents every year, and Intel says many semi-structured and unstructured documents are routed manually. AI can automate the processing and categorization of these documents for faster routing and reduced labor costs.

Using a Support Vector Classification (SVC) model, this kit was optimized with the Intel distribution of Modin and the Intel extension for Scikit-learn powered by oneAPI. Intel claims these tools improve data preprocessing, training, and inference times to be 46%, 96%, and 60% faster, respectively, compared to the stock implementation of the Index Kit. Accenture’s intelligent document system without Intel’s optimizations for reviewing and sorting documents at 65% accuracy.

Over the next year, Intel announces that it will release a series of additional open-source AI benchmark kits with machine learning and deep learning models trained to help organizations of all sizes on their journey. of digital transformation.


About Author

Comments are closed.