Ublion learns, some context

Your are here:
< Back

Ublion Document AI is able to perform tasks that normally require human intelligence, such as visual perception of invoice content, invoice type recognition, language recognition, country recognition, and conversion. It has the ability to recognize patterns, it learns from invoices, and it becomes more intelligent over time without any programming needed. It independently changes models and learns what to look for. Ublion recognizes all type of invoices, as AR, AP and credit notes.

Ublion processes visual information from invoices from which it identifies specific objects. This means that Ublion customers don’t have to map invoice fields nor have to create templates to get their invoices recognized. Invoices are just send to the Ublion machines and the deep learning model does to work detecting and recognizing invoice content and type from your invoice. This is all done in real-time and the result is populated in an electronic invoice format.

As Ublion improves on it’s own, one of the main challenges is that no customer will get a perfect recognised invoice every time. So, our main consideration of the Ublion service is what level of quality—in precision, recall, latency, or another aspect—should be required for Ublion? We decided that an average recognition of 82% should be the main baseline. Customers can increase their recognition performance by training the Ublion with their invoices. Therefore we introduced auto-training. With customer models, training modules and pipelines we offer an average invoice recognition of 92% – 99% and an average of 30 seconds invoice processing time for our public cloud customers.

Appealing examples about AI neural networks

Google Search. Google started its search engine with PageRank, which used dozens or hundreds of factors as part of an algorithm which determined the order in which Google displayed results. PageRank was essentially a strict set of rules set by humans. Developers periodically tweaked the algorithm to boost the importance of certain ranking factors. But in 2015, the company began using an additional layer called RankBrain, which uses a neural network to help determine search results. The product now improves on its own, meaning that Google has given up some control over the way its product works to AI. This is how Ublion works too.

Excel. Imagine that you are writing a simple formula in Excel. Every time new numbers (data) are introduced to that formula, it calculates an answer. Excel is not asked to do anything else other than calculate an answer based on the fixed math of that formula. Now imagine that you introduce a very large data set to that formula and instruct that formula to look for combinations and patterns and then learn from that data. Based on this process, the formula begins to change and learn what to look for and how to make that calculation better and quicker. How did the formula change? Why did it choose to calculate that data using a different mathematical approach? That’s the part that is in a “black box”. This is how Ublion works too.

AI vs. Machine Learning vs. Deep Learning

AI, machine learning and deep learning are three terms often used interchangeably to describe how smart the software is. And especially AI is used as an umbrella term for any computer program that does something smart. However, to understand how Ublion works it is useful to understand the key distinctions among them. Therefore, think about deep learning, machine learning and artificial intelligence as a set of Russian dolls nested within each other, beginning with the smallest and working out. Deep learning is a subset of machine learning, and machine learning is a subset of AI. In other words, all machine learning is AI, but not all AI is machine learning, and so forth.


Most AI solutions are using a lot of of if-then statements, offered as a statistical model mapping. For example to map invoice data to categories. The if-then statements are simply rules explicitly programmed by a human hand. We call this legacy AI.

Machine learning

Machine learning is a subset of AI. One aspect that separates machine learning from AI is its ability to modify itself when exposed to more data. Machine learning does not require human intervention to make certain changes. The “learning” part of machine learning means that algorithms attempt to optimize along a certain dimension. Each algorithm has an predefined objective, controlled by humans.

Deep learning – Ublion

Deep learning gives Ublion the ability to learn from invoices without being explicitly programmed.

Deep learning is a subset of machine learning and refers to the number of layers in a neural network. These multiple layers allow deep neural networks to automatically create features when it learns of the data. Without any human intervention. It creates a so-called feature hierarchy, because simple features (for example two pixels) recombine from one layer to the next, to form more complex features (for example a line). In character recognition, features may include histograms counting the number of black pixels along horizontal and vertical directions, number of internal holes, stroke detection and many others.

Copyright - Ublion is powered by Get Data Insight - www.getdatainsight.com