Core ML

Core ML. It is a machine learning framework developed by Apple Inc. for integrating trained models into applications running on its ecosystem of devices. The framework is designed to be highly efficient, allowing models to run directly on the device's processor, including the A-series, M-series, and GPU, enabling on-device inference for privacy and performance. This enables developers to add features like image recognition, natural language processing, and predictive analytics to apps for iPhone, iPad, Mac, Apple Watch, and Apple TV.

Overview

Core ML serves as the foundational machine learning runtime across Apple's operating systems, optimized to leverage the company's custom silicon. It is tightly integrated with other Apple frameworks, such as Vision for computer vision tasks and Natural Language for text analysis, providing higher-level APIs for common use cases. The framework supports a wide variety of model types, including neural networks, tree ensembles, and support vector machines, which are converted into its proprietary format. By executing models on-device, it ensures user data does not need to leave the device, aligning with Apple's strong stance on user privacy and data security.

Features and capabilities

A key feature is its support for on-device execution, which minimizes latency, allows functionality without a network connection, and protects user data. It utilizes hardware accelerators like the Neural Engine, a dedicated processor block in modern Apple silicon, to perform matrix multiplications and other operations with high energy efficiency. The framework supports fine-tuning or transfer learning on-device with tools like Create ML, allowing models to be personalized with user data. For complex pipelines, models can be chained together or combined with custom preprocessing code written using Accelerate or Metal for maximum performance.

Integration and deployment

Developers integrate trained models into an Xcode project simply by dragging the converted model file into the project navigator; Xcode then automatically generates a Swift or Objective-C programming interface for it. For deployment, the model is bundled within the application's IPA or .app package, making it instantly available when the app is installed from the App Store. Real-time performance is achieved by leveraging the entire hardware stack, from the CPU and GPU to the Neural Engine, with the system automatically selecting the best hardware for a given task. This integration is central to many system features in iOS and macOS, such as those in Photos and Siri.

Model conversion and tools

Models trained in popular frameworks like TensorFlow, PyTorch, and scikit-learn must be converted to the Core ML model format using tools such as Core ML Tools, a Python package created by Apple. The conversion process often involves simplifying model architectures and quantizing parameters to reduce size and improve inference speed on mobile and edge devices. Create ML, an application within Xcode, provides a graphical interface for training and exporting models using data from Swift Playgrounds or datasets prepared in macOS. For advanced optimization, developers can use the Core ML Model Optimization toolkit to apply techniques like pruning and palettization.

History and development

The framework was first announced at the WWDC in 2017 as a unification and evolution of earlier, more specialized inference engines used within Apple. Its development has been closely tied to the advancement of Apple silicon, with each new generation of the Neural Engine bringing substantial improvements in supported operations and model complexity. Major updates have been introduced at subsequent WWDC events, expanding support for training on-device, new model types, and enhanced tools like Create ML. The ongoing development reflects Apple's strategic focus on enabling powerful, private AI capabilities across its entire product lineup.

Category:Apple Inc. software Category:Machine learning frameworks Category:IOS software Category:MacOS software