The need

Machine learning models benefit from transfer learning. In computer vision, neural networks trained on a large dataset are useful for initialising models on other vision tasks. How can we leverage the transfer learning technique for text?

The idea

Suppose you have a very domain specific question and you want to get the best answer from a document. What if we could read thousands of documents and then use AI to answer that question with the best context and inference?

The solution

Using a neural network called the Reasoning Network (ReasoNet), researchers mimic the inference process of human readers. With a question in mind, ReasoNet scans a document and focuses on different parts until it gets an answer.

Technical details of Machine Reading

MRC requires modelling complex interactions between the context and the query. Using a novel neural network architecture called the Reasoning Network (ReasoNet), researchers were able to mimic the inference process of human readers.

With a question in mind, ReasoNet reads a document repeatedly, each time focusing on different parts of the document until a satisfying answer is found or formed. Microsoft researchers today have been able to surpass human-level parity on SQuAD dataset using a unique MRC algorithm called R-NET: Machine reading comprehension with self-matching networks. R-NET applies a self-matching attention mechanism to refine the representation by matching the passage against itself, which effectively encodes information from the whole passage.

When we applied these MRC algorithms to the book, Future Computed by Brad Smith and Harry Shum, it was incredible to see that we can answer so many interesting questions. We can apply this to solve enterprise data challenges and answer enterprise domain specific questions.


Projects related to Machine Reading

Browse more innovation sandbox projects

Explore the possibilities of AI

Jump-start your own AI innovations with learning resources and development solutions from Microsoft AI.