Machine Learning with CoreML

Arturo Arriaga
Jan 3, 2023
2 min read

This project explores ML/AI development within the iOS ecosystem. This product is an iOS app that performs image classification of a user drawings. It contains a machine learning model trained on Apple’s CreateML framework, takes a drawing performed by the user and classifies it into one of 10 classes. To use the app, users draw a doodle and submit the drawing for classification. The app uses the trained machine learning model, called ArriagaDrawingClassifier, to analyze the drawing and provide a label for the object in the drawing.

Technologies Used Core ML is the framework developed by Apple that allows developers to integrate machine learning models into their iOS, iPadOS, macOS, watchOS, and tvOS applications. Core ML provides a unified representation for all models, allowing them to be easily integrated into applications and run on a variety of different devices. The framework is imported as module withing the Xcode environment is use to import a fully trained machine learning model.

Create ML is an application developed by Apple that allows developers to create and train machine learning models. This is the platform used to train the model ArriagaDrawingClassifier. Create ML is very different from the development peformed in the course. One main difference is that in CreateML, developers can use a simple, drag-and-drop interface to build, train, and evaluate machine learning models. Little in-depth knowledge of machine learning algorithms or techniques is needed. Create ML supports a variety of different machine learning tasks, including image classification, object detection, and natural language processing. While I found this interface to be very easy to use. The primary drawback is in being able to customize your model and instead you are only given very view options for optimizing your model.

Dataset The data used for this project comes from the Quick Draw Dataset collected and open sourced by Google for the game “Quick, Draw!” While the entire dataset contains over 50 million drawings across 345 categories, this project simplifies this into 10 categories and about 10,000 png images collected for training this model. The game and dataset, and its corresponding GitHub repo can be explored by following the next two links: