At Serimag we consider the implementation of different projects through the application of the various Artificial Intelligence technologies that we know about to be as important as investment in R&D, in order to continue to develop this technology. And it is the latter that concerns us now, being something which we are especially proud of.

In each project evaluation and in each production process Serimag incorporates the most appropriate technology to the needs of both the service and our client thanks to the professionalism and knowledge of all the technical staff and project managers. We live in a rapidly changing technological world in which new techniques are continually being developed to improve existing processes. At Serimag we want to be part of this technological revolution.

Cooperation with the Barcelona Supercomputing Center (BSC)

After months of joint research between Serimag and the Barcelona Supercomputing Center (BSC-UPC), a new single graphics and text classification system based on neural networks has been developed which will be promoted by Serimag in all our document classification processes.

Traditional document classification systems based on rigid, predefined templates and rules were surpassed some years ago by highly automated production environments such as those offered by TAAD. However, standard automation based on document intelligence continues to process documents separately with respect to graphics and text, merging both aspects in a final classification system:

Previous classification - image and text

Although these classification systems are based on Machine Learning, they require intermediate and explicit layers between one processing and another, which inevitably prevent the maximum benefit from intelligent recognition engines and require manual parameterisation with consequent maintenance.

Serimag started a cooperation project with BSC a year ago based on the following hypothesis: there must be a neural network architecture capable of performing complete classification (of both graphics and text) in an integrated manner and without the need for parametric coupling modules, understanding the document as a whole.

Neural networks classification

Today we can say that the hypothesis was true and that it does actually offer improvements in terms of production time and results.

Application in projects with Tessi-Graddo

At Serimag we offer a constantly evolving service. During the life of a process we periodically evaluate potential technological improvements coming from our R&D laboratory that improve automation rates to benefit both the client and Serimag. With three years of joint cooperation in automated document processing, at Serimag we have started to introduce this new technology in the production processes that we share with Tessi-Graddo.

The new model of neural networks has been implemented in current production models together with Tessi-Graddo, managing to reduce the rejection of previous models by half and increasing accuracy to above 99%:

Increasing automation

We are already working on new systems that are also applicable to document segmentation and data extraction, which we hope to be able to incorporate into production during 2019. Results such as these demonstrate our commitment to R&D, cooperation with the leading research centres in their areas and the impact it has on continuing to offer a better service to our clients.