Data Mining and Intelligent Systems Research Group
MIDUSI is a research group at the III-LIDI institute of the Facultad de Informática, Universidad Nacional de La Plata, in La Plata, Argentina. We work on sign language translation, interpretable computer vision, learning from data streams and machine learning on microcontrollers.
Our current project, Inteligencia de Datos. Técnicas y Modelos de Machine Learning (2023–2026), is directed by Waldo Hasperué and co-directed by Laura Lanzarini. The group has 9 members.
Our work has 3,262 citations on Google Scholar and a group h-index of 22. The figure combines all 9 members' profiles, counting co-authored papers once, as of 2 November 2026. Profiles with more than 100 papers are only partly listed, so their remaining citations are added as they are.
Research
Computer Vision and Sign Language. Deep learning for Argentinian Sign Language recognition and translation from video and poses, interpretability of vision models (including medical imaging), digitisation of historical astronomical plates and analysis of satellite imagery. We released the LSA16 and LSA64 datasets.
Intelligent Data Analysis and Big Data. Learning from data streams with concept drift, metaheuristics for feature selection, machine learning on large data volumes, time series forecasting with deep learning, and applications to health, education and finance.
TinyML and Edge AI. Efficient neural networks for microcontrollers and low-power devices: model compression, quantization and deployment, including our EmbedIA framework.
Text Mining and Language Models. Keyword extraction, summarization and topic modelling, and language-model-based assistants for university life.
Datasets
- LSA64: A Dataset for Argentinian Sign Language. 3200 videos of 64 signs of Argentinian Sign Language performed by 10 subjects, 5 repetitions each.
- LSA16: A Handshape Dataset for the Argentinian Sign Language. 800 images of 16 handshapes of Argentinian Sign Language, performed 5 times by 10 subjects.
- LSA-T: A Continuous Argentinian Sign Language Dataset. About 22 hours of video from the CN Sordos YouTube channel, split into clips with their Spanish subtitles and the signers' keypoints, for sign language translation.