ValgrAI continues its commitment to disseminating knowledge in the field of artificial intelligence through its series of Research Mornings. These events, which have become a reference point for the scientific and technological community, offer a virtual space for the exchange of ideas and the presentation of cutting-edge research in the field of AI.
The twelfth edition of the ValgrAI Research Morning maintains the virtual format through the Teams platform, adapting to current needs and allowing for broader and more diverse participation. This format has proven effective in connecting researchers, academics, and professionals from different locations, fostering enriching dialogue on the latest trends and developments in artificial intelligence.
The event is structured around two main presentations, each addressing different but equally relevant aspects in the field of AI. These presentations not only offer a detailed view of specific research but also provide a broader context on how these advances can impact various sectors and society in general.
Below is a summary of the two presentations that formed the core of this XII ValgrAI Research Morning Meeting:
Dynamic brain state assignment using echo state networks
Dr. Wael el-Deredy presented research focused on the use of Echo State Networks (ESN) for dynamic brain state assignment. This research is based on the premise that transient and recurrent activity patterns in spontaneous EEG are fundamental computational properties of the brain.
Dr. el-Deredy’s work explores how ESNs, a specific type of recurrent neural networks, can be used to capture and analyze these brain patterns. ESNs are characterized by randomly connected neurons and function as non-linear dynamic systems, making them particularly suitable for this type of analysis.
The main objective of this research is to provide a compact representation that reflects the evolution of brain signals over time, reproducing their non-linear dynamics. This approach could have significant implications for understanding brain function and developing more advanced brain-computer interfaces.
Identification and categorization of racial stereotypes in texts and sexist memes
The second presentation, given by Elias Urios Alacreu, focused on the development of deep learning systems for the identification and categorization of racial stereotypes in texts and sexist stereotypes in memes.
This research, which is part of a master’s thesis, proposes a novel approach based on the learning-by-disagreement paradigm. Instead of seeking consensus among annotators, this method integrates different perspectives into the systems, aiming to make them more generalizable.
Urios Alacreu presented results suggesting that, in some cases, this paradigm outperforms the classical approach. Additionally, his work includes an analysis of the importance of context in stereotype detection, highlighting its effectiveness particularly in meme classification.
This research has potential implications for online content moderation and the creation of more inclusive and respectful digital spaces.
