University of Science and Technology of China (USTC) researchers say they have built a deep-learning machine that outperforms the average human's ability to answer verbal reasoning questions.
Traditionally, computer science researchers have used data-mining techniques to analyze huge volumes of texts to find the links between words they contain; this approach assumes each word has a single meaning represented by a single vector. The USTC researchers solved this problem by taking each word and seeking other words that often appear nearby in a large corpus of text, and then using an algorithm to see how these words are clustered. They then looked up the different meanings of a word in a dictionary and matched the clusters to each meaning, with the overall result being a way of recognizing the different senses that some words can have.
The researchers also developed a method to make it easier for a computer to answer verbal reasoning questions by identifying the category of each question so the computer then knows which answering strategy it should employ. The researchers then developed an algorithm for solving each one using the standard vector methods.
From Technology Review
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found