Causal Probabilistic Based Variational Autoencoders Capable of Handling Noisy Inputs Using Fuzzy Logic Rules

Abstract

Researchers and engineers may use inferential logic and/or fuzzy logic to solve real-world causal problems. Inferential logic uses probability theories, while fuzzy logic uses its membership functions and set theories to process uncertainty and fuzziness of the events. To benefit from both logics, some researchers in the past tried to create probabilistic fuzzy logic (PFL). Deep Learning algorithms (DLs) with their incredible achievements such as very high precision results in some specific tasks are at the center of the weak AI. However, DLs fail when it comes to causal reasoning. In order to equip Deep Learning algorithms (DLs) with reasoning capabilities, one solution would be to integrate non-classical logics such as PFL with DLs. In this paper, we will demonstrate the first step toward creating a deep causal probabilistic fuzzy logic architecture capable of reasoning with missing or noisy datasets. To do so, the architecture uses fuzzy theories, probabilistic theories, and deep learning algorithms such as causal effect variational autoencoders.