MIT’s ‘Gen’ Aims To Make AI Techniques More Accessible

MIT has unveiled an artificial intelligence system that it said could make an array of artificial intelligence techniques more accessible to programmers, while also offering a helping hand to experts in the field.

Researchers said the system, called Gen, is similar to TensorFlow, a set of tools developed by Google for automating AI tasks, principally those involved with deep learning and neural networks.

Gen takes a similar approach, but allows developers to make use of a broader array of AI techniques, including probabilistic programming, MIT said.

Some probabilistic programming systems allow the flexibility to use several kinds of AI techniques, but run inefficiently compared to Gen, the developers said, saying they aimed to combine automation, flexibility and speed.

Ease of use

“If we do that, maybe we can help democratise this much broader collection of modelling and inference algorithms, like TensorFlow did for deep learning,” said Vikash K. Mansinghka, one of the MIT researchers involved with the project.

The name “Gen” comes from the researchers’ aim to make the tool practical for general-purpose use.

Gen was presented at the Programming Language Design and Implementation conference last week.

It allows users from fields including computer vision, robotics and statistics to write models and algorithms without having to deal with equations or manually write high-performance code.

It also allows AI experts to write sophisticated models and inference algorithms, used for prediction tasks, that would previously have been unfeasible.

In their paper, for instance, the researchers demonstrate how a short Gen program can infer 3D body poses, a difficult computer-vision inference task.

Expert appeal

The program includes components for graphics rendering, deep learning and different types of probability simulations.

Gen is simple enough to be used by anyone from novices and experts, the researchers said.

“One motivation of this work is to make automated AI more accessible to people with less expertise in computer science or math,” said Marco Cusumano-Towner, a PhD student in MIT’s Department of Electrical Engineering and Computer Science and one of the authors of the paper.

“We also want to increase productivity, which means making it easier for experts to rapidly iterate and prototype their AI systems.”

Gen is already being used by Intel in a collaboration with MIT and by an MIT-IBM Watson AI Lab project, amongst others.