Uber AI Labs is Uber's artificial-intelligence lab. Researchers there built their own programming language for AI applications, and are releasing it for anyone to use.
For a company founded by Travis Kalanick, popular with its ride-hailing empire and hard-nosed business tactics, this is generous of them to hand out their in-house innovations to anyone including their potential competitors.
"It's more important for us to engage with the whole community instead of keeping it ourselves," said Zoubin Ghahramani, the chief scientist at Uber’s AI Labs.
As a ride-hailing business with many users, Uber’s goal of bringing reliable transportation to everyone requires effortless prediction and optimization. The company is relying a lot on its AI-mediated guesswork to sort things out.
Like for example, suggesting optimal routes, deciding where demand for rides will be, finding sensible pool combination, knowing what pool passengers should be put into, and even creating the next generation of intelligent vehicles.
Uber's AI is called Pyro. With expertise of data scientists, engineers, and other users, it's built to improve the ways the company can analyze those parameters. It uses a technique known as deep probabilistic modeling, a mixture of two AI methods: deep learning and Bayesian modeling.
Combining two AIs makes Pyro a specialist at dealing with uncertainty, much as calculus is the mathematics for reasoning about rates of change. Models that are built in the language of probability can capture complex reasoning, know what they do not know, and uncover structure in data without supervision.
Pyro for example, can guess where demands will be high in the future by learning previous information, like where cars were needed on a given day in the past.
The key principles that form the basis of Pyro’s design are:
- Universal: Represent any computable probability distribution.
- Scalable: Scales to large data sets with little overhead.
- Minimal: Implemented with a small core of powerful, composable abstractions.
- Flexible: Aims for automation when you want it, control when you need it.
Built on the PyTorch library, which includes automatic differentiation using very fast, GPU-accelerated tensor math. PyTorch constructs gradients dynamically, which enables Pyro programs to include stochastic control structure. Random choices in a Pyro program can control the presence of other random choices in the program.
Whis way, Pyro can represent any probabilistic model, while providing automatic optimization-based inference that is flexible and scalable.
In Pyro, both the generative models and the inference guides can include deep neural networks as components.
The resulting deep probabilistic models have shown great promise in recent work, especially for unsupervised and semi-supervised machine learning problems.
Ghahramani says his colleagues want to be full participants in the research community, taking part in conferences and presenting papers, and that Uber’s AI lab aims to release more open-source projects in the future.
Being active in the larger research community is, of course, an advantage in recruiting more human power with knowledge in the field.
Even Apple that is famous for being secretive, is publishing research papers to attract talents. And Uber in releasing its AI for the public to play with, is a matter of self-interest as much as it is a gift to the AI community.