1 - Artificial-Intelligence System Surfs Web to Improve Its Performance
Information extraction involves classifying data items that are stored in plain text, and is a major area of research for machine learning scientists. Last week, a research team from MIT introduced a new approach to information extraction for machine learning systems at the Association for Computational Linguistics’ Conference on Empirical Methods on Natural Language Processing, and won a best-paper award. Instead of feeding their system as much data as possible, the team's winning approach takes a different route and focuses on a much smaller data set, a similar process used by human beings - if you're reading a paper that you don't understand, you're likely to do a search on the web and find articles that you are able to understand. This new system approach does something similar; if the system's confidence score is low in assessing a particular text, it will query for more information, pulling up a handful of new articles from the web that correlate with a specific set of terms. In future, this model could be applied to sparse data and save much time in reviewing databases.