Core ML
Updates to the Core ML framework bring even faster model loading and inference. The new Async Prediction API simplifies the creation of interactive ML-powered experiences and aids in maximizing hardware utilization. Use the new Core ML Tools optimization module to help compress and optimize your models for deployment on Apple hardware. Weight pruning, quantization, and palletization utilities can be applied during model conversion or while training your model in frameworks like PyTorch to preserve accuracy during compression.
Create ML
Use the Create ML app or framework to build custom models on top of Apple’s latest visual feature extractors for images and multilingual transformer-based embeddings for text. Creating a model to understand the content of images has never been easier with the addition of multilabel classification, interactive model evaluation, and new APIs for custom training data augmentations.
Machine learning
New APIs in the Vision framework provide advanced image segmentation, animal body pose detection, and 3D human body pose leveraging depth information. Use VisionKit to easily integrate Visual Lookup and subject-lifting experiences into your app. The Natural Language framework enhances understanding of multilingual text using new transformer-based embedding models. The Speech framework makes it easy to introduce custom vocabulary for speech recognition, so you can personalize your user experiences.