Uber Introduces New Platform for Deep Learning, No Programming Required

Deep learning is a field of artificial intelligence that involves machines learning how to do tasks and make decisions that humans would normally be required to do. It is a branch of machine learning. The term deep learning comes from the way algorithms are designed to adapt and learn from information in a similar way deep neural networks in a human’s brain enable learning to happen.

Many companies have used deep learning research to provide better products and service. Uber has been one of those companies actively working on deep learning models. According to Uber, this has been for a variety of reasons some of which include customer support, object detection, improving maps, and preventing fraud.

To accomplish this, Uber has built and improved upon an already existing vast array of open source tools. Beginning in 2017 Uber released two new tools that enabled further deep learning research. One of which was Pyro, an open source probabilistic programming language. The other was Horovod, an AI tool that delegated the deep learning tasks throughout an array of different machines and GPUs. These both utilized currently existing open source libraries including TensorFlow, PyTorch, CNTK, MXNET, and Chainer.

Uber is now announcing their next contribution to the open source deep learning toolbox with Ludwig. Ludwig is a deep learning program focused on making the process easier to understand for both, the in-experienced and experienced researchers. This is achieved through five design principles that are at the foundation of Ludwig.

First is users can train a model and use it for predictions without coding. Second, the tool is usable across many different use cases through a principle called generality. The third is giving users extensive control over model building and training, in other words flexibility. Fourth is extensibility through the ease of adding new models and features. Last up is what Uber calls understandability, a principle that should allow users to visualize and perceive performance easier.

Uber claims Ludwig can reduce hours of coding down to just a few minutes of work. Through an input as simple as a CSV file and Ludwig’s easy to use encoder-decoder architecture, a wide array of machine learning tasks can be accomplished including text classification, object classification, image captioning, regression, and language modeling to name a few. In addition, several visualization options are available for when the results are collected.

We are now in a time where machines can learn to solve complex problems without the help humans. Although this may sound scary, great strides can be made in technology through deep learning medicinal and pharmaceutical advances, virtual assistants, and an area Uber may be able to reap benefits from in the future: vision for driverless cars.