Amazon's AutoGluon: Generating AI Models with Just 3 Lines of Code

2020-01-12

Building machine learning applications that include image, text, and tabular datasets is not an easy task. It requires feature engineering, which involves using domain knowledge to create features that work well with AI algorithms, as well as extensive dataset preprocessing to ensure that the trained model does not have any biases. This is why Amazon developed AutoGluon, aiming to enable developers to write intelligent applications with just a few lines of code. AutoGluon's goal is to automate many decisions that developers previously had to make themselves. Tasks like hyperparameter tuning, which are typically performed manually, require scientists to predict how different choices made during the construction of an AI model will affect its training. There are also complex engineering tasks, such as neural architecture search, where developers must determine the best design for their respective models, which can be time-consuming and labor-intensive. AutoGluon can automatically tune known tasks within default ranges, providing well-performing options with just a few lines of code. Developers only need to specify when the trained model is ready, and in response, AutoGluon will utilize available computing resources to find the strongest model within the allocated runtime. It builds upon Gluon, which was developed by Amazon and Microsoft three years ago. Gluon is a machine learning interface that allows developers to build models using a collection of pre-built and optimized components. AutoGluon, on the other hand, is an end-to-end development process. AutoGluon can immediately recognize models for table prediction, image and text classification, and object detection. It provides an API that experienced developers can leverage to further enhance the predictive performance of their models. It requires Python 3.6 or 3.7 and currently only supports Linux, but Amazon has stated that Mac OSX and Windows versions will be released soon. AWS applied scientist Jonas Mueller stated in a statement: "We developed AutoGluon to democratize machine learning and make it accessible to all developers. The end result is impressive, as AutoGluon addresses the aforementioned issues by automatically adjusting all options within default ranges, known for its good performance on specific tasks and models."

RECENT AI NEWS

RECENT AI TOOLS