Students at the Massachusetts Institute of Technology have successfully created a new artificial intelligence programming language. They say Gen can be used easily by anyone, from novices to experts. Without complex code, this novel probabilistic programming system allows users to write complex models and reasoning programs for statistical predictions, making them more accessible to all.
Many accurate prediction models used to require a lot of manual coding, but Gen has changed that. Gen can both lead novices into THE field of AI and help experts make new advances, allowing them to prototype their ideas and write code for their own automated AI systems with just a few lines of code.
Researchers are trying to combine the best attributes of AI, such as automation, flexibility and efficiency. “If we can do this, maybe we can help democratize a broader range of modeling and reasoning algorithms, much like TensorFlow has done for deep learning,” argues Vikash K. Mansinghka, a member of the team that developed Gen.
Where does Gen apply? According to the MIT paper, Gen can be applied to model writing and algorithms in a variety of AI technologies, such as computer vision, robotics and statistics, without having to deal with equations or manually write high-performance code.
A short Gen program helps users infer difficult computer vision reasoning tasks, such as infering body positions in 3D. This has applications in autonomous systems, human-computer interaction and augmented reality.
Not only that, Gen programs contain components that perform graphical rendering, deep learning, and probabilistic simulation types. The combination of these different technologies is more accurate and faster than earlier systems developed by some researchers.
Gen can simplify data analysis by using another Gen program that automatically generates complex statistical models typically used by experts to analyze, interpret, and predict the underlying patterns in data. Earlier systems required a lot of manual coding to make accurate predictions.
What makes Gen different? Unlike deep learning platforms like TensorFlow, PyTorch, and Theano, Gen programs explicitly break down modeling and reasoning.
By automating the process to calculate the proposal density required for various advanced Monte Carlo techniques, Gen provides an excellent platform for the combination of Julia and TensorFlow code.
Gen is far superior to existing probabilistic programming languages in solving reasoning, including 3D body poses for single-depth images, the paper notes. Gen can also provide a high-level infrastructure for reasoning tasks using methods such as optimization, variational reasoning, certain probabilistic methods, and deep learning. Gen has more flexible reasoning programming capabilities that make performance improvements possible.
As an example, Gen has found its niche in the following direction: Intel and MIT are working on a depth-sensing camera for Gen’s augmented reality system.
MIT’s LincolnLaboratory is applying Gen to aero-robotics for humanitarian relief and disaster response. Gen is at the heart of the mit-ibm Watson AI Lab project being conducted by the Defense Advanced Research ProjectsAgency, which aims to mimic the level of human general knowledge of an 18-month-old baby.
Uber’s chief scientist, vice president of artificial intelligence and professor at the university of Cambridge ZoubinGhahramani, who was not involved in the research, said, “Gen represents a major advance in the field that will contribute to scalability and practical AI system implementation based on probabilism.”
Peter Norvig, Director of research at Google, who was not involved in the study, also has high praise for Gen: “[Gen] allows people with problems to use probabilistic programming to come up with more principled solutions to the problems themselves, and is not constrained by the choices of probabilistic programming system designers. General-purpose programming languages succeed because they make it easier for programmers to get things done, while also allowing programmers to create entirely new things to solve new problems efficiently. Gen does the same for probabilistic programming.”