The computer engineer Mikel Ferrero Jaurrieta (Pamplona, 1995) developed new methods for “improve the way artificial intelligence (AI) processes information” in order to make it “more sensitive to word order and thus able to better understand the context and nuances of texts”.
This improves your ability to complete tasks. How to identify whether the tone of a piece of writing is positive or negativedetermines whether a message is “spam” or not and automatically classifies different types of technical documents, as explained in his doctoral thesis, defended at the Public University of Navarra (UPNA).
To function “efficiently”, the artificial intelligence It depends on information processing, the fundamental tasks of which are comparison, classification and fusion. These tasks “become more complex when the data has multiple dimensions or factors.”
THE thesis of Mikel Ferrero proposes new procedures “to manage this complexity, allowing computers to sort, compare and merge information more efficiently by taking into account how different parts of the data interact with each other.”
Concretely, the researcher applies new methods of information fusion in a field of artificial intelligence: neural networks, which are computer systems inspired by the functioning of the human brain.
Specifically, “there is a challenge” in word processing: “the importance of the order of words or sentences, especially when there is a temporal dependence between them”, according to Ferrero.
For example, in a text, a word can have a different meaning depending on those which precede or follow it (thus, in the sentence “not good”, this last word loses its positive character, because of those which precede it), and this is “crucial to understand the message”.
WORD ORDER MATTERS
In order to meet “this challenge”, Michael Ferrero developed new non-symmetric information fusion operators. In conventional neural networks, symmetric operators are used, “which have difficulty grasping the importance of the order between words”.
On the contrary, non-symmetric methods for combine information from different parts of the text “consider order to matter, because they do not treat all word combinations the same way.
These methods allow neural networks analyze the text taking into account the importance of word order and how they relate to each other over time.
This “improves the ability of computers to perform tasks such as identifying the emotional tone of text, determining whether a message is spam, and classifying different types of documents such as patents or questions.”
THE doctoral thesis of Mikel Ferrero was led by Carlos Lopez Molinaprofessor at the Department of Statistics, Informatics and Mathematics at UPNA, and Zdenko Takác, professor at the Slovak University of Technology in Bratislava (Slovakia).
Furthermore, the the research received funding of the public company Tracasa Instrumental, the National Research Agency and the European Regional Development Fund (ERDF).
Mikel Ferrero holds a degree in computer engineering from UPNA (2018) and Master in Modeling and Mathematical Research, Statistics and Computer Science from the Public Universities of Navarra, Zaragoza and Basque Country (2020).
Throughout his professional career he has worked at Tesicnor SL, Hiberus Tecnología, National Center for Technology and Food Safety (CNTA), UPNA (Documentary Fund of Historical Memory of Navarre) and Instrumental worries.
During his doctoral thesis, carried out within the research group of Artificial Intelligence and Approximate Reasoning (GIARA)has published six articles in journals considered prestigious in the field of computer science and artificial intelligence such as “Engineering Applications of Artificial Intelligence”, “IEEE Transactions on Fuzzy Systems” or “Information Fusion” and received two awards at international conferences. . He is also co-author of fifteen papers at national and international conferences.