Each March, Silicon Valley heads into conference season. Tech behemoths like Google, Facebook, Intel and Nvidia welcome thousands of developers to their tech showcases — always with a deep focus on artificial intelligence. Today it was Apple’s turn to unveil new machine learning tools and software at its 2018 Worldwide Developers Conference (WWDC), which runs June 4-8 at the McEnery Convention Center in San Jose, California.
Apple announced Create ML, a new machine learning tool that enables developers to easily import custom data and create computer vision and natural language AI models on the macOS.
“Create ML is designed to let you train without being a machine learning expert,” said Craig Federighi, Apple Senior Vice President of Software Engineering.
Create ML is built on Swift, Apple’s general-purpose programming language for its operating systems and Linux. Developers can use the Xcode Playgrounds UI to train models by simply dragging training data into a default model builder.
Create ML speeds up model training on external GPUs. In the past, Mac users could only run AI training with in-laptop Intel CPUs and GPUs or by renting computational power from cloud service vendors such as AWS or Google Cloud. Last week, Apple officially added support for external GPUs to macOS, powering computation in games, video editing, and AI processing.
Apple’s Metal 2, a technology released last year to optimize performance, graphics and computation for GPUs, can also accelerate neural network training on macOS.
With Create ML, developers can save significant time on training AI models. Federighi says the training time of an object detection model based on a 20,000-image dataset is reduced from 24 hours to just 48 minutes on MacBook Pro.
It’s not only Apple, other tech giants are also striving to simplify developers’ entry into the world of AI. Microsoft’s Custom Vision and Google’s AutoML can also create sophisticated models with minimum effort and knowledge.
Core ML 2
Last year, Apple released Core ML, a technology for high performance on-device machine learning applied to Apple products such as Siri, Camera, and QuickType. Core ML helps developers integrate disparate machine learning models into apps. It supports extensive deep learning models with 30+ layers, along with standard tree ensembles, SVMs, and generalized linear models.
Today, Apple released Core ML 2, a better-then-ever iteration that can process 30 percent faster using batch prediction, and reduce model size by 75 percent using quantization.
Apple’s Increasing Ambitions in AI
Unlike many of its Silicon Valley neighbors, Apple is a relative latecomer in this field. Apple’s researchers had not released any machine learning papers until late 2016, and its smart speaker HomePod which went on sale this January is experiencing sluggish sales.
In a bid to catch up, Apple poached Google Search and Artificial Intelligence head John Giannandrea in April to lead its machine learning and AI strategy efforts.
Apple today announced major updates for Siri, including ShortCut, a feature that allows iOS users to create custom Siri voice commands for third-party apps integration. When ordering coffee at Starbucks for example, a user can preset a voice command for “a grande ice latte with soy milk and one sugar” and simply voice activate with the word “latte.”
While Amazon Alexa and Google Assistant are leading the smart assistant battle in the US, Siri’s ShortCut is a smart, practical and cost-effective move by Apple that builds on its forte in the mobile ecosystem to explore the potential of voice commands.
Apple is also investing in the autonomous vehicle market. Last month the company registered 55 self-driving cars with the California Department of Motor Vehicles, which makes it the state’s second largest such fleet.
Apple has about 20 million developers on its AppStore, which has become the world’s largest app marketplace with a staggering 500 million visitors per week. Developers surely appreciate Apple simplifying the transfer of AI features into popular applications. Apple may be trailing in the AI race but it has the resources to close the distance quickly.
Journalist: Tony Peng | Editor: Michael Sarazen