A group of researchers from MIT described a new probabilistic programming system called “Gen” in a paper presented at the Programming Language Design and Implementation conference last week. Without having to deal with equations or manually write high-performance code, users write models and algorithms from multiple fields where AI techniques are applied. Gen’s difference comes into play at this point, allowing experts to write complex models and inference algorithms.

For example, in their article, the researchers show that a short Gen program can take on a difficult computational inference task involving 3D models, autonomous system applications, human-machine interactions, and augmented reality. Behind the scenes, this program includes components that perform types of graphics rendering, deep learning, and probability simulation. Thanks to the combination of these various techniques, it provides better accuracy and speed than previous systems developed by some researchers.

Due to its simplicity, researchers say that Gen can be easily used by anyone from novices to experts. Marco Cusumano-Towner, a PhD student in the Department of Electrical Engineering and Computer Science, says: “One of the motivations for this work is to make automated AI training more accessible to people with less expertise in computer science or mathematics,” and adds, “We also want to increase productivity; this means making it easier for experts to rapidly iterate and prototype artificial intelligence systems.”

Source: news.mit.edu/2019/ai-programming-gen-0626