Ai startup 9
ML - The way the world works - analyzing how things work - A podcast by David Nishimoto
Categories:
deepset works to get more meaningful search results. Deepset uses transfer learning, language models, and question and answer to drive search results. Making sense of text data. deepset is an open source company. It uses natural language processing to answer questions using bert. MiniLM is a small version of BERT. deepset uses extractive questioning and answering where the answer must be part of the text by one span. This is an energy problem. There is a large set of potential predictions. SQuAD has become the default dataset. BERT is a feedforward network predicting the start and end token of the extractive answer then maps to the word dictionary to produce and answer. Industry wants question and answer for its data. There is a growing number of knowledge workers spending hours and hours reading text. Knowledge workers are use to web search and enterprise search. Scaling involves gathering a large number of documents. The road map include generative question and answering and sythentic reasoning.