Deep Learning for Machine Reading Comprehension

Established: September 1, 2016



The goal of this project is to teach a computer to read and answer general questions pertaining to a document. We recently released a large scale MRC dataset, MS MARCO.  We developed a ReasoNet  model to mimic the inference process of human readers. With a question in mind, ReasoNets read a document repeatedly, each time focusing on different parts of the document until a satisfying answer is found or formed. The extension of ReasoNet (ReasoNet-Memory) incorporates the shared memory component in the model has been applied on Knowledge Graph Completition Task. We also develop a a two-stage synthesis network (SynNet)  for transfer learning in machine reading comprehension. Our latest MRC model, called SAN (Stochastic Answer Net), simulates multi-step reasoning using stochastic prediction dropout, achieving state-of-the-art on SQuAD.