By Eric Xing, CEO, Founder, and Chief Scientist at Petuum
AI as a concept is not unfamiliar — most of us are aware of some amazing examples of AI technology, including some that have outsmarted humans in certain competitions. But what really makes AI attractive is its potential to become part of our workflows to dramatically improve our productivity.
There are already examples of what AI solutions have to offer: in robotics, AI-enabled robots can perform difficult and dangerous operations; in healthcare, AI tools can speed-up and improve diagnoses; in energy and environmental applications, AI processes can reduce overhead and waste; and in manufacturing, AI tools can improve quality and predict maintenance needs.
Despite all of these promising applications, only a small fraction of companies have tapped into AI to transform their businesses. This is because under the shiny surface of what seems to be a simple solution, there’s a huge body of backend dirty work that needs to be done.
In order to maintain and customize even the simplest implementation, a team of engineers must complete tasks such as data wrangling, feature engineering, model compilation, algorithm design, distributed training, debugging, resource provisioning, hardware management, fault recovery, and many others. Even for a resource-rich company, it remains difficult and time-consuming to build and maintain in-house AI solutions. AI continues to require very specialized talent and intensive infrastructure.
However, high-quality off-the-shelf solutions are not currently available. If you can find a plug-and-play AI solution, it’s usually just a primitive prototype with very limited flexibility, customizability, stability, compatibility, and services.
We want to change this.
At Petuum, we have tasked ourselves with industrializing AI technology. We aim to completely redefine the key elements of a successful AI technology. We are building AI components that are as intuitive and accessible for businesses as bricks, cranes, nuts, and bolts are for building houses. We are creating standardized and reusable building blocks for sustainable AI solutions that can be built quickly and economically. To that end, our system does not depend on any particular assumptions about data form or format, meaning businesses can deploy our AI components with any type of raw data in any environment, from large-scale datacenters to public clouds, to a single workstation, and from mobile devices to cutting-edge IOT.
By providing businesses with a new way to approach AI, we are enabling them to build, manage, and maintain their own AI solutions easily, without needing to invest tons of time and capital into recruiting talent and building up infrastructure. This new engineering approach is both timely and imperative to fundamentally disrupt today’s elaborate, fragile, and expensive AI.
In future blog posts we’ll share some examples of how our clients in various industries including medical, industrial, financial services, and more, are using our AI building blocks to create sustainable high-quality AI solutions.