Anything new with machine learning this year?

Yes, there are several exciting updates related to machine learning this year at WWDC. Here are some of the key highlights:

1. Apple Silicon and Training Models:

   • Apple Silicon's hardware acceleration and unified memory are being leveraged for training models using popular frameworks like PyTorch, TensorFlow, JAX, and MLX. This allows for efficient training and deployment of machine learning models on Mac devices (Platforms State of the Union).

2. Machine Learning Frameworks:

   • Apple's built-in machine learning frameworks have been enhanced with new capabilities. This includes APIs for natural language processing, sound analysis, speech understanding, and vision intelligence. The Vision framework, in particular, is getting a new Swift API this year (Platforms State of the Union).

3. CreateML Enhancements:

   • The CreateML app now includes an object tracking template for training reference objects to anchor spatial experiences on visionOS. Additionally, new time series classification and forecasting components are available for integration within apps (Explore machine learning on Apple platforms).

4. Performance Improvements for Transformer Models:

   • Significant performance improvements have been made for transformer models, including support for 8-bit and 4-bit integer quantization, fused scale dot product attention, and unified memory support. These updates help in running larger models more efficiently on Apple devices (Train your machine learning and AI models on Apple GPUs).

5. Apple Intelligence:

   • Apple introduced "Apple Intelligence," a personal intelligence system that brings powerful generative models to iOS, iPadOS, and macOS. This system enhances the ability to understand and generate language and images, helping users take actions with rich awareness of personal context (Platforms State of the Union).

6. Running Models on Device:

   • Apple has made it easier to run a wide array of models on their devices, including large language models and diffusion models. This includes models like Whisper, Stable Diffusion, and Mistral, which can be fine-tuned and optimized for on-device performance (Explore machine learning on Apple platforms).

For more detailed information, you can check out the sessions:

• Platforms State of the Union
• Explore machine learning on Apple platforms
• Train your machine learning and AI models on Apple GPUs