Machine Learning with Apple’s Core ML 3 is Exciting and Personal

by Jun 22, 2019#MachineLearning, #FrontPage, #iOS

Printer Icon

The scope of machine learning is just beginning to be imagined. The number of applications has increased at a humongous rate in the last years. Nowadays, almost all activity that includes data user analysis relies in machine learning. Fields like medicine, sports or even arts are taking advantage of apps using machine learning to improve the current procedures.

For some time Apple has provided a framework to take advantage of this powerful tool with Create ML, Core ML, and some Domain APIs. Create ML permits to create and train custom models. With Core ML the developers can integrate machine learning models into theirs apps. Core ML has always supported diverse models, e.g., Generalized Linear Models, Tree Ensembles, Support Vector Machines, and FeedForward, Convolution, and Recurrent Neural Networks. Domain APIs includes functionality for image, speech, and sound  analysis.

Create ML has added new model types to provide a total of 9 types:

  • Image classifier.
  • Object detector.
  • Sound classifier.
  • Activity classifier.
  • Text classifier.
  • Word tagger.
  • Tabular classifier.
  • Tabular regressor.
  • Recommender.

For this year Apple has introduced new Domain APIs that expands the reach of the framework. Sentiment detection allows to classify text in real time according to its positive or negative nature. Word embedding provides the ability to find semantically similar words. For example “Moon” is semantically close to “Night”  and far from “Dog”. A new fully embedded speech to text converter that not only analyze what is spoken but how is it, making possible to differentiate from a normal voice from a high jitter.

In the WWDC19 was presented Core ML 3, which is optimized for on-device performance and ensures the privacy by doing all the process locally and not running in any server. The newest version of Core ML provides  model flexibility and model personalization. Now, Core ML has support for more than 100 Neural Network layers allowing to import the state of the art models into an app.  A new converter from TensorFlow is already in place and a ONNX converter is soon to be released. The model gallery has also been updated. Core ML 3 allows On-device Model personalization. This personalization reflects the capability to adjust and tune the model on the device. One single model in an app can be adjusted for each user for his personal use.

Although Core ML will be supported on all Apple devices (iPad, iPhone and iWatch), one important consideration is that Core ML is made exclusively for iOS operating system. This is a big limitation against Google ML kit, that works for Android and iOS.

Photo by Wahid Khene on Unsplash

About Us: Krasamo is a mobile development and consulting company focused on the Internet-of-Things and Digital Transformation.

Click here to learn more about our mobile development services.


Android App Development 101

Android App Development 101

Native Android app development strategies currently revolve around Android’s commitment to becoming Kotlin-first and their migration from Java to Kotlin. In this transition, businesses are looking for first-class support, interoperability, and lower development and maintenance costs.

9 Tips to Help You Become a Successful Android App Developer

9 Tips to Help You Become a Successful Android App Developer

Over 60% of mobile phones worldwide run on Android. Being an Android app developer can be a profitable career if you do it right. Here are eight tips to help you land your app at the top, instead of getting lost at the bottom of the barrel and became a successful Android app developer.

Kotlin 1.3 is an Exciting Option for Android Mobile Apps Development

Kotlin 1.3 is an Exciting Option for Android Mobile Apps Development

Kotlin has been adopted by many Android developers and companies because it offers a concise programming syntax which makes developers more comfortable writing code, prevents the common errors seen when developing in Java, and is easy to switch from iOS development given that Kotlin syntax is very similar to Swift.