What’s new in MagicaLCore 2.0.1: - Help at every step: tap Training, Validation, Loss or Memory to see what they mean, with an example chart. There is also help about the training data and the augmentations, and you can read it while a model is training. - More reliable validation: each training always checks itself with the same samples, also when you resume it, and the exported model is the one from its best moment. - Clearer pause: while it stops, the screen tells you and keeps everything locked. - Testing: the live camera recognizes hands better in any orientation, hand action videos open fine even when they are in iCloud, and the import button opens Files the first time. - Tabular data: testing from the History compares the prediction with the real value, and training again always works. - Free plan: exporting and testing your models always works, even if an ad can’t load. - Updated tutorials and more improvements for VoiceOver and larger text. - Bug fixes and stability improvements.

MagicaLCore 2.0 is the biggest update yet. - Six kinds of models: images, sounds, tabular data, text, hand gestures and hand movements in video. - Test your models instantly with photos and files, or live with the camera and microphone. - Larger, more reliable training, even with data on an external drive. - Pause training and pick it up later from the history, right where it stopped. - Tutorials for every model type, with good practices for preparing data. - A new Liquid Glass design, light and dark mode, and better VoiceOver accessibility. - Available in 13 languages, including Arabic. - Export models to use them in Swift Playgrounds and Xcode. Thanks for using MagicaLCore. If you enjoy it, a rating helps a lot.

V1.1.0: • Fixed a bug that caused the interface to temporarily freeze during model training • Fixed a bug that changed the interface size in the tutorial • Fixed a bug where selected parameters were kept in a new training session • Added German and Traditional Chinese to the list of available languages

Version 1.0.1 – What’s New This update introduces advanced new controls for model training, offering greater customization and precision: New Features • Dropout Probability: Adjust how many neurons are deactivated during training to help prevent overfitting. • Learning Rate: Controls how quickly the model updates its parameters. • Early Stopping Tolerance: Defines the minimum improvement threshold required to continue training. • Patience (No-Improvement Iterations): Sets how many consecutive epochs without improvement are allowed before training stops. Improvements • Visual and functional enhancements to controls, with explanatory popovers and intelligent range validation.