Intelligent use of household appliances for energy consumption optimization (T7)

Research challenges / Novelty / Innovation
• Identification of AI-based strategies for the intelligent use of household appliances to optimize energy consumption.
• Processing heterogeneous data, learning from imbalanced data, lack of sufficient relevant data, different levels of intermittency in processing, and time series prediction.

Research results:
• Prediction of future usage.
• Functional prototype.
• Personalized recommendations.

Innovation:
• Design and implementation of machine learning-based methods to address data complexities.
• Improvement of energy resource management strategies at the household level.

Multilinguality, multi-task learning, imbalance in NLP (T1)

Provocări în cercetare / Noutate / Inovare
Provocări: Modelele bazate pe transformatoare nu au capacități reale de raționament , lipsa de date etichetate în alte limbi decât engleza, provocarea adaptării modelelor la domenii noi, transferul lingvistic și cultural, Conținutul generat de modele generative poate conține biasuri sociale, costuri ridicate ale metodelor post-antrenare
Noutate:
• Dezvoltarea de instrumente lingvistice pentru sarcini de înțelegere a limbajului natural, prin tehnici eficiente de particularizare a modelelor mari de limbă
• Dezvoltarea de tehnici de evaluare, identificare si mitigare a biasurilor sociale în modelele generative de text si multimodale
• Dezvoltarea de metode eficiente și sigure de antrenare a modelelor de limbaj, care să fie robuste la atacuri adversariale, și eficiente din punctul de vedere al utilizării resurselor
• Dezvoltarea de metode de interpretabilitate și tehnici de raționare pentru modele text-imagine sau text-video.

Research results:
• Methods, techniques and technologies resulting from the research topics
• Demonstrator for the application theme

Inovation:
• Automation of human tasks through natural language understanding
• Increasing trust in text-based AI applications through interpretability, robustness, and bias reduction

Brain activity analysis using AI techniques (T5)

Research challenges / Novelty / Innovation
Challenges: variability of biological signals, noise, EEG signal is non-stationary, high dimensionality of features, clustering quality
• Techniques for identifying brain activity
• Feature representation
• Computational methods for characterizing brain activity
• Integrated solutions for understanding brain activity

Research results:
• Advanced AI models for brain signal processing and analysis (feature extraction, feature representation, clustering/classification problems)
• Demonstrator that adapts and integrates proposed models into a specific use case

Inovation:
• Development of customized AI models for brain signal processing and analysis.
• Integration of innovative models into applied systems in the medical field (e.g. diagnostic systems)