Field of Research Artificial Intelligence

We are bridging the gap between academic research and practical applications of Artificial Intelligence (AI) by addressing fundamentals, quality assurance, and approaches to overcome practical limitations. Our activities enable AI to be standardized and implemented in multiple sectors.

Fundamentals

• Novel algorithms: methods to handle small  data sets, multimodality, and missing data 
• Customized architectures: deep learning architectures for specific applications (e.g., incorporation of ancillary Information)
• Spectrum of models: implementation of deep learning, Bayesian inference, autoencoders, supervised/unsupervised learning, and recurrent models
• Theoretical considerations: contributions to the basic understanding of AI model development and use (e.g., problem-solving behaviors) 
• Integration of domain knowledge  (e.g., models, sparsity, and distributions)
• Assurance of dependability and security  (e.g., privacy preservation).e standardized and implemented in multiple sectors

Quality assurance

• Explainable AI: methods to visualize,  explain, and interpret deep learning models 
• Trustworthiness: methods to account for  noisiness, outliers, and nonstationarity 
• Robustness: methods to defend from  adversarial attacks 
• Uncertainty: methods to quantify  uncertainty of model outputs 
• Evaluation: measures of data and  model quality
• Online Machine Learning with good  tracking capabilities

Overcoming practical limitations

• Model size reduction: approaches for  neural network compression
• Efficient inference: training and  implementation of models despite limited  storage, computation, and energy resources
• Federated learning: method for several participants to jointly train models while  allowing data to remain at the site of collection
• Distributed learning: methods to adapt to communication constraints
• Low complexity: solutions that offer low computational and energy cost 
• Extremely low latency via massive parallelization