INTELIGENT SYSTEMS

PI:   João Leite

• The Intelligent Systems (IS) group is a national leader in AI research, with a strong international reputation in Knowledge Representation and Reasoning (KRR) and Machine Learning (ML). The IS group, which includes an EurAI fellow, results from the realignment of the Knowledge-Based Systems group to face current challenges of AI, through addition of experts in ML. Members regularly publish in top journals e.g., AIJ*, JAIR, TNNLS*, and Bioinformatics* and top conferences e.g., IJCAI*, AAAI*, KR*, ICML*, AAMAS*, and ISWC, and belong to their PCs.
  
  Recent achievements of AI relied on scaling data sizes, model capacity, and compute power. As we exhaust human-generated data, progress towards robust, trustworthy AI requires the exploitation of domain knowledge in ML, the extraction of symbolic actionable knowledge from ML models, and merging data-based learning with knowledge-based reasoning. Towards this long-term vision, our research questions include:
  
  How can we develop models whose components can mechanistically encode domain knowledge in training and inference as inductive biases, such as physical constraints or domain-elicited knowledge such as rules, so that the model has just enough capacity to learn each task?
• How can we develop sound ways to help humans interpret current deep models, or design newmodels that humans can interpret at different scales? How can they be safely trusted? How can we generate knowledge from these models and effectively use it in combination with existing domain knowledge? What knowledge representation languages and reasoning procedures are required?
  
  How can the path that S4 models took be enhanced and their key components be identified, leading to more efficient sequence learners? How can these new models be used in automated decision-making, particularly in reinforcement learning?
  
  These RQs are being addressed on several projects, with a focus on application to Bioinformatics, Aerospace, Territory Management, and Robotics.