Google recently introduced AdaNet, which is an open-source tool designed for integrating machine learning algorithms in a bid to attain better predictive insights. The tool is currently available on the Tensor GitHub storage area.
“AdaNet builds on our recent reinforcement learning and evolutionary-based AutoML efforts to be fast and flexible while providing learning guarantees,” Google AI software engineer Charles Weill said in a blog post. “Importantly, AdaNet provides a general framework for not only learning a neural network architecture but also for learning to ensemble to obtain even better models.”
AdaNet leverages a technique dubbed ensemble learning to help it in combining and improving algorithms. According to Charles Weil, a Google AI software engineer, this approach previously called for too much training time or domain expertise.
To make the implementation of AdaNet easier, the framework connects to the TensorFlow Estimator to deliver important information to one place. It also plugs into the TensorBoard, which aids in delivering visual feedback when an artificial intelligence (AI) model is undergoing training.
The recently introduced open-source tool by Google ensures that learning guarantees for the models that it builds not only through learning neural networks’ architecture but also including subnetworks to them.
Machine learning experts who desire to gain additional control of this process can take advantage of TensorFlow APIs, especially in defining their subnetworks, toggling other settings, and customizing loss functions.
Extra details regarding how AdaNet operates can be found in the published paper that was presented at the International Conference on Machine Learning last year.
The recent launch of AdaNet marks the newest step forward in AutoML, which is Google’ way of automating the training and deployment of neural networks. Google Cloud Platform launched AutoML for natural language processing (NLP), computer vision, translation and Cloud Auto ML intended for creating custom artificial intelligence (AI) models later in January.