Publication

Challenge 1: Integrating knowledge, reasoning and learning

Book Chapter (2021)

Book Title

Artificial Intelligence, Robotics & Data Science

Publisher

Consejo Superior de Investigaciones Científicas

Pages

19-37

Volume

11

Serie

CSIC scientific challenges: Towards 2030

Doc link

http://libros.csic.es/product_info.php?products_id=1493&PHPSESSID=8afc5bcf66efab7725ce14484c2eff79

File

Download the digital copy of the doc pdf document

Authors

  • Manyà Serres, Felip

  • Colomé Figueras, Adrià

  • García de Polavieja, Gonzalo

  • Ríos Insua, David

  • Torres Barrán, Alberto

  • Armengol i Voltas, Eva

  • Blum, Christian

  • Flaminio, Tommaso

  • Godo Lacasa, Lluís

  • Levy, Jordi

  • Meseguer González, Pedro

  • Segovia Aguas, Javier

Abstract

Knowledge, reasoning and learning (KRL) play a central role in artificial intelligence (AI) and are instrumental in solving many AI complex problems. In such problems we may have massive amounts of data and imprecise models, and the goal is to create AI systems that scale well in such scenarios, which often requires combining KRL techniques. In this chapter, we first present KRL from an historical perspective and then identify future research directions in the KRL domain in which CSIC is or could be very competitive. In particular, we present challenges integrating KRL from several perspectives, such as:


General planning : finding AI agents that combine learning and planning techniques to be able to learn solutions to a problem and generalize them to others.


Problem solving : create algorithms with learning capabilities for solving complex optimization problems with huge amounts of data.


Learning adaptable AI agents from a reduced amount of data, keeping the most information possible from data, while being computationally and sample efficient for adapting to changing situations.


Enhancing logics for conditional, causal reasoning and their integration with machine learning (ML).


Integrating uncertainty, similarities, knowledge and learning by using stochastic methods that better represent the variability in real-world scenarios.


These challenges are considered key topics in the current AI research and, with CSIC already having a good knowledge on them, pushing research in those directions could place CSIC as a reference institution in the world on these fields. We detail a plan and resources needed to boost research on the integration of KRL techniques.

Categories

artificial intelligence, knowledge engineering, learning (artificial intelligence), planning (artificial intelligence), uncertainty handling.

Scientific reference

F. Manyà, A. Colomé, G. García, D. Ríos, A. Torres, E. Armengol, C. Blum, T. Flaminio, L. Godo, J. Levy, P. Meseguer and J. Segovia. Challenge 1: Integrating knowledge, reasoning and learning. In Artificial Intelligence, Robotics & Data Science, 19-37. Consejo Superior de Investigaciones Científicas, 2021.