Create ML

Create ML
Name	Create ML
Developer	Apple Inc.
Released	2018
Latest release	2024
Operating system	macOS
License	Proprietary
Website	apple.com

Create ML Create ML is a machine learning model development tool for macOS designed to enable on-device training and export of models for Apple's platforms. It integrates with Apple's Xcode ecosystem and the Core ML model format to streamline workflows for developers, designers, and researchers targeting iOS, iPadOS, macOS, and watchOS. The tool emphasizes ease of use through graphical and programmatic interfaces while leveraging hardware and software advances from Apple.

Overview

Create ML provides a set of training APIs and a graphical app to create supervised and unsupervised models that can be converted to Core ML for deployment. It ties into Apple technologies such as Metal (API), Swift (programming language), and macOS system frameworks to accelerate training and inference. The tool complements cloud-based services and research toolkits by focusing on privacy-preserving, on-device model creation, aligning with initiatives exemplified by Secure Enclave and corporate privacy policies.

History and development

Create ML was announced during an Apple Worldwide Developers Conference keynote as part of Apple's push to make machine learning accessible to app developers. Its evolution parallels releases of Xcode updates, macOS versions, and expansions in Core ML capabilities. Early iterations emphasized symbol classification and image labeling; subsequent updates added support for text, tabular data, and custom model export, coinciding with advances showcased at events like WWDC 2019 and WWDC 2020. Partnerships and ecosystem growth involving companies that integrate App Store apps have influenced real-world adoption, while hardware developments such as Apple silicon chips (e.g., M1 (Apple silicon), M2 (Apple silicon)) increased on-device training feasibility.

Features and capabilities

Create ML offers features for dataset preprocessing, model training, evaluation, and export. It supports image classification, object detection, sound classification, text classification, and tabular regression and classification tasks, aligning with formats used by Core ML. The framework exposes training hyperparameters and evaluation metrics like accuracy and F1 score, enabling experimentation similar to workflows found in frameworks such as TensorFlow and PyTorch. Integration with Metal Performance Shaders allows hardware acceleration on Apple GPUs and NPUs, while Swift-based APIs permit automation and reproducibility within Xcode projects. The tool can export models in formats optimized for deployment to devices sold through the App Store.

Workflow and user interface

Create ML can be used via a graphical Create ML app or programmatically through Swift packages and Playgrounds within Xcode. The graphical interface guides users through dataset import, label mapping, training configuration, and live evaluation with visual metrics and confusion matrices. The programmatic APIs enable scripting of experiments, dataset augmentation, and integration with continuous integration systems used in professional app development workflows maintained by teams publishing to the App Store Connect. Model artifacts are typically tested within simulators and on-device using Xcode debugging tools before submission to app distribution channels governed by App Store Review Guidelines.

Supported models and frameworks

Create ML produces models compatible with Core ML and leverages on-device primitives via Metal (API) and Accelerate (Apple framework). It does not directly produce models in formats native to frameworks like TensorFlow or PyTorch but supports interoperability by enabling conversion paths through tooling in the broader machine learning ecosystem. Supported model categories include convolutional classifiers, transfer learning from pre-trained vision backbones, object detectors, sound classifiers built on audio features, and tabular models using gradient-boosted decision trees and linear models influenced by paradigms found in libraries such as XGBoost and scikit-learn.

Use cases and applications

Developers and organizations use Create ML to build features like image-based search in retail apps distributed via the App Store, voice or sound recognition in accessibility tools aligned with iOS accessibility initiatives, personalized recommendation models for media apps, and document classification for enterprise workflows that integrate with Apple Business Manager. Academic and industry researchers prototyping on-device privacy-preserving classifiers have adopted Create ML when targeting Apple hardware demonstrations at conferences such as NeurIPS and ICML. Education programs teaching practical app-integrated machine learning frequently incorporate Create ML into curricula that include Swift Playgrounds and mobile development modules.

Limitations and criticism

Create ML faces criticism for being tightly coupled to Apple's ecosystem, limiting portability to non-Apple platforms and constraining collaboration with teams using server-side frameworks like TensorFlow and PyTorch. Observers note that advanced model research requiring bespoke architectures, large-scale distributed training, or specialized layers remains better served by research-grade toolkits used at institutions such as MIT and Stanford University. Limited transparency around some optimization internals and dependence on proprietary hardware acceleration have prompted concerns similar to debates seen with other platform-specific AI offerings; critics compare these trade-offs to open alternatives prominent in communities like OpenAI and open-source initiatives led by organizations such as the Linux Foundation. Despite these caveats, Create ML is valued for rapid prototyping and streamlined deployment workflows to Apple platforms.

Category:Apple software