Do you still need to scan some paper docs into your claims portal? Then this is news baby, things just got smarter.
Omniscience today announced it has been granted a patent for technology that creates a breakthrough engine for Optical Character Recognition (OCR). Using innovative machine learning technology enables products based on this patent to handle a broader array of forms than more generic extraction tools such as those from Amazon, Google, or others.
This Omniscience technology uses many learning strata to rapidly extract and digitize hand-written forms and tabular data from virtually any source, regardless of its complexity. It is first being applied to difficult problems in insurance underwriting and, over time, will be made available for business process outsourcing (BPO), financial, national security, and other applications. Because of the speed and accuracy of the results of this advanced technology, complex decisions can be handled with much less risk.
The patent is called “Systems and Methods for Machine Learning based Content Extraction from Document Images.” Manu Shukla, Omniscience co-founder and Chief Technology Officer explained the fundamental breakthrough: There is no limit to how many machine learning strata that can stack for data extraction, but each additional stratum makes coordination much harder; Omniscience has addressed this through algorithmic distribution, which allows vast amounts of data to be processed rapidly and completely.
Examples of problems that can be deciphered by this technology include low-resolution, faxed or copied data in character-based Asian languages like Japanese and Chinese. These challenges appear often in many industries. Currently, this OCR Engine technology is being used in Omniscience Customer Intelligence products which handle material such as hand-written health and personal forms for insurance applications, reports from many different labs and in many languages, insurance treaties, claims, credit card receipts printed on thermal paper, credit card images, IDs, among others.
Explaining how this technology is different from what has been available to date, Shukla said, “More generic extraction tools from companies like Amazon, Google, or others are not able to scale to different types of forms in a template-agnostic way and keep accuracy very high.” This means that most digitization technologies end up with a human in the loop and a high need for data correction along the way. Omniscience focused on the harder problem of digitizing forms and materials without prior-known templates, and without humans in the loop.
“With all of our technology and product development, the goal is to push the limits of both applied and original machine learning and artificial intelligence to speed enterprises to faster, more accurate, and ultimately more useful business — regardless of the complexity or volume of data and data sources,” said Shukla. “In addition to the technology in the OCR engine patent, we are currently also working on several types of innovations: AI that enables ubiquitous parallelization of deep learning and leads to very fast processing of many layers of complexity and AI for new kinds of simulations of catastrophe modeling including wildfire danger and risk assessment for capital management and other financial markets.”