Arya.ai launches Braid to integrate 'deep learning' into systems

Artificial Intelligence (AI) startup Arya.ai has announced the global launch of Braid, an open-source meta-framework that is designed to help researchers speed up the process of building complex Deep Neural Networks, a method of building intelligence into systems. According to the company, building neural networks, that function like human brains with complex layers, is a resource-intensive, expensive and time consuming process. While other developer platforms and tools exist to develop neural networks, the startup claims that Braid has an added advantage of being flexible, customizable, and modular, that works with operating systems for Deep Learning. The brainchild of two Indian Institute of Technology Bombay alumni Vinay Kumar Sankararapu, and Deekshith Marla, Braid is designed to be scalable for use with networks that need the handling of many data points at large volume. Illustrating with an example, Sankarapu, who is the CEO and Founder of Arya.ai, said that for AI to understand, say, medical radiology reports, a neural network with millions of neurones trained and defined into the architecture would be needed to learn the task.

“Currently, developing applications on these frameworks needs substantial resource investments in terms of specialized skills and long cycle times. But now, with Braid, it is easy to start up development immediately and deploy the tool,” explained Kumar.

Arya.ai claims to have designed the new tool for rapid development and to support arbitrary network designs allowing for quick experimentation without having to worry about the boilerplate components of the code. With Braid, developers can not only customize but also add to existing neural layers with simple coding. Braid is currently made available free and open source, to companies that use research scientists and technologists developing these networks.

“Open sourcing key tools in AI, will help discover newer, interesting and more impactful use cases and applications for AI that we may not have even thought of,” says Kumar.