This paper introduces a method to learn structured embeddings of Knowledge Bases (KBs) using a neural network architecture that maps symbolic representations into a continuous vector space. The goal is to enable the use of KB data in various AI applications such as natural language processing, computer vision, and collaborative filtering. The proposed method learns embeddings for entities and operators for relations, allowing for flexible and efficient representation of structured data. The embeddings are trained to preserve the original knowledge while enabling new inferences. The method is tested on WordNet and Freebase, and it is shown to be effective in predicting relations and entities. Additionally, the method is adapted for knowledge extraction from raw text. The paper also discusses the use of Kernel Density Estimation (KDE) to estimate probability densities in the embedding space, which helps in improving the accuracy of predictions. The results show that the method performs well in ranking and generalization tasks, and it can be extended to other AI applications. The approach is flexible, compact, and capable of generalization, making it suitable for large KBs. The paper concludes that the method provides a promising way to leverage structured data for various AI tasks.This paper introduces a method to learn structured embeddings of Knowledge Bases (KBs) using a neural network architecture that maps symbolic representations into a continuous vector space. The goal is to enable the use of KB data in various AI applications such as natural language processing, computer vision, and collaborative filtering. The proposed method learns embeddings for entities and operators for relations, allowing for flexible and efficient representation of structured data. The embeddings are trained to preserve the original knowledge while enabling new inferences. The method is tested on WordNet and Freebase, and it is shown to be effective in predicting relations and entities. Additionally, the method is adapted for knowledge extraction from raw text. The paper also discusses the use of Kernel Density Estimation (KDE) to estimate probability densities in the embedding space, which helps in improving the accuracy of predictions. The results show that the method performs well in ranking and generalization tasks, and it can be extended to other AI applications. The approach is flexible, compact, and capable of generalization, making it suitable for large KBs. The paper concludes that the method provides a promising way to leverage structured data for various AI tasks.