Turn meanings to numbers. Overview and application of semantic embedding

Natural Language has always been a tough kind of input for machines to be exploited. Along with human evolution, the way we express ourselves has developed many complex features and approximate rules that can’t be measured or quantified. Nowadays companies trying to process huge amounts of human generated word or sentences in a systematic way have found a powerful weapon in deep learning technique of embeddings, that can turn them into numbers that well represent the same characteristics in terms of syntax and meanings. In this speech we will explain how companies like Google, Amazon, IBM and Microsoft have made huge efforts in creating the most efficient and reliable models to achieve the state of the art in embedding techniques and how their application has changed many tools we commonly use. Then we will explain how Machine Learning Reply has applied in first person this amazing models to a real life use case.