My PhD project

My PhD was started in October of 2020 as an opportunity-driven project. I began reading about Dempster Shafer Theory and these readings led me to questions related with evidence, uncertainty, belief and plausibility.

How should evidence be stored? As individual pieces of evidence? As their combination? Is this hard computation always worth?

Those questions soon reached the multi agent paradigm: During a communication process between agents, what should be shared? Individual pieces of evidence, merged evidence, belief or knowledge? Actually, would an agent understand other’s evidence? What conventions should they have in common?

All these questions point out different areas to study: evidence representation, reasoning with uncertainty, computational complexity, belief revision and multi agent systems.

Little by little, we transformed these ideas in a concrete situation where the interconnections among these areas can be explored. Given two agents A₁ and A₂ with Θ₁ and Θ₂ their domain of knowledge respectively, and a set of evidence about subsets of their domains each one; we are (openly) interested in the following lines:

1. Stablishing a common domain between Θ₁ and Θ₂. In the context of Dempster Shafer Theory, there are already some related works (here and here), but we would also like to give a definition independent of this theory, for example, attending to the logics of the individual domains.

2. Defining strong (enough) arguments to filter evidence in order to share it with another agent attending to time and other parameters such as receiver’s goals or background (for instance, being able to avoid redundant or “boring” information). These filters could also depend on properties of the proper evidence, such as uncertainty and contradictions. Here, we propose two different approaches: one more theoretical, studying the problem from some modal logics perspective; and another one supported by data experiments.

As a result, we could define a protocol for the interchange of the most relevant evidence between agents in polynomial time.

Once this would be done, our learner agent (the receiver) could process this new evidence and enlarge its knowledge, change its belief or just forget it (as we do as learners many times). Anyway, this is another issue.

As you can see, I am involved in a very exciting and interdisciplinary project. At this point, I would like to read, listen and discuss as much as possible, so don’t hesitate to contact me to make me suggestions or just to talk!