
handle: 10578/23171
Aunque actualmente sean tendencia y estén siendo muy utilizadas, las redes neuronales no tuvieron un fuerte impacto cuando fueron inicialmente propuestas. Su estado de implantación actual se debe principalmente al gran bene cio que aportan en multitud de problemas. Su gran crecimiento en los últimos tiempos se ha debido a la conectividad y propagación de las redes en torno a lo que hoy en día se conoce como Internet, que produce una cantidad de datos masiva. Pero sobre todo se debe al desarrollo de la tecnología, y a la existencia de hardware con gran capacidad de computo. Esta combinación de avances informáticos, junto con la gran cantidad de datos disponibles, o que se puedan minar, ha conseguido convertir a las redes neuronales en uno de los algoritmos de deep learning más populares hasta la fecha. Este proyecto en concreto, forma parte de un convenio entre la universidad y la compañía Indra, empresa que en la actualidad está desarrollando una línea de investigación dentro del campo del deep learning. De alguna forma, el problema de Indra es relativo a la documentación, necesitan automatizar la gestión documental relativa a facturas del departamento Imaging. Este departamento se encarga de resolver problemas de una forma clásica, ya bien sea utilizando técnicas OCR en documentos estructurados, o búsqueda de patrones estáticos en elementos que siempre son iguales en su identi cación. El problema planteado por la empresa Indra Sistemas versa sobre la necesidad de construir un clasi cador multiclase general para clasi car automáticamente documentos, utilizando técnicas de machine learning, deep learning y natural lenguaje processing. El siguiente documento, pretende exponer un sistema que incluye una serie de técnicas, modelos y resultados que resuelven el problema proporcionado por Indra Sistemas
Although currently trendy and widely used, neural networks did not have a strong impact when initially proposed. Their current state of implementation is mainly due to the great bene t they bring in a multitude of problems. Their great growth in recent times has been due to the connectivity and spread of networks around what is now known as the Internet, which produces a massive amount of data. But above all it is due to the development of technology, and the existence of hardware with large computing capacity. This combination of computer advances, together with the large amount of data available, or that can be texted, has managed to turn neural networks into one of the most popular deep learning algorithms to date. This speci c project is part of an agreement between the university and Indra, a company that is currently developing a line of research in the eld of deep learning. Somehow, Indra’s problem is related to documentation, they need to automate the document management related to invoices of the Imaging department. This department is in charge of solving problems in a classical way, either using OCR techniques in structured documents, or searching for static patterns in elements that are always the same in their identi cation. The problem raised by Indra Sistemas is the need to build a general multiclass classi er to automatically classify documents, using machine learning, deep learning and natural language processing techniques. The following document aims to present a system that includes a series of techniques, models and results that solve the problem provided by Indra, and integrate the system into a web application written with the Django framework. una aplicación web escrita con el framework Django.
Informática, Aplicación informática
Informática, Aplicación informática
| selected citations These citations are derived from selected sources. This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | 0 | |
| popularity This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network. | Average | |
| influence This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically). | Average | |
| impulse This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network. | Average |
