The Artificial Intelligence Research Institute (IIIA) offers two scholarships for the introducción to a research career in the context of CSIC's JAE Intro programme.

Projects and mentors:

1. Management of ontologies and folksonomies for defining education competencies (JAEIntro-2020-IIIA-02):
   Mentors: Carles Sierra, Marco Schorlemmer

   Education competencies are normally defined by the education departments of governments. The definition of these competencies is a process not without discrepancies and disagreements among the members of the educational community. These competencies define the educational profile of a student, who achieve them by passing certain modules and subjects. However, the set of competencies (from now on ‘tags’) is relatively general. For example, you can include the tag ‘To know how to program', but not ‘To know how to program in Python'. Also, the competency requirements of companies not only use the tags as defined by the government’s education department, but they use very different tags at times. For example, ‘To know how to manage a human team', and furthermore, this set of tags often evolves faster than the tags used in formal education. All this causes alignment problems between job applications and educational profiles. An example is the difficulty in assigning students in the “FP dual” programme as the tags of the applications and those of the students do not match well. This work would focus on defining functionalities that allow managing the ontologies that allow defining student profiles. In particular, to allow teachers and employers to add and blend new tags to an ontology (à-la-folksonomy), to define matching functions between students and job offers that take into account the relationships between tags that can be defined by all (hierarchies, meronomies, ...), to define optimal educational paths that allow to achieve the sets of competencies required by industrial training, etc. The work will be to build a webapp to be used by the education department of the government of Catalonia in the tasks associated with “FP dual” training. A pilot use of the program is planned for a small set of schools.

2. Reinforcement learning for automating NPC behaviour in simulation environments (JAEIntro-2020-IIIA-03):
   Mentors: Jordi Sabater-Mir, Josep Lluís Arcos.
   Normally, entities not controlled by the user (NPCs) in simulation or video game environments base their behaviour on "scripts" that define how they must react to world events given their internal state. This mechanism is very laborious and requires the stage designer to foresee all the possibilities in advance. It is starting to be common the use of “reinforcement learning” (specifically “Deep reinforcement learning”) to generate behaviour in a much more automatic way and obtaining much better and more robust results. Game engines such as Unity are beginning to incorporate tools to train NPCs using "reinforcement learning" (ml-agents) and, in conferences such as SIGGraph more and more examples of the use of this technology are presented. This project will consist in the application of this technology in emergency simulation environments to automate the behaviour of NPCs that appear in a 3D scenario (whether they are civilians, mobile units or emergency personnel). Relevant knowledge: Courses in Artificial Intelligence (especially knowledge in “reinforcement learning”; advanced knowledge of Python programming; Experience with Unity.