AIAI Seminar - Monday 24 January 2022 - Talk by Ruiqi Zhu

Title:  

Thesis Proposal:  Continuous Probabilistic Knowledge Base Evolution

Abstract:

Knowledge base(KB) is an ideal vehicle to tackle many challenges. In recent years, research in knowledge-based applications has been prosperous, such as question answering, recommender systems, Root Cause Analysis, etc. Nevertheless, most knowledge base construction methods are error-prone, so probabilities and other uncertainty measures are used to quantify potential mistakes. Despite that, the real world is changing, so the knowledge could become outdated. We demand methods to update the probabilistic knowledge and repair the tacit faults.

Our proposal aims to devise a method that automates the probabilistic knowledge base evolution so that 1) probabilities of knowledge can match the degrees of beliefs of expert humans w.r.t the changing world; 2) outdated and contradictory knowledge can be revised so that the KB can remain consistent and up to date.