Update README

Author: Hrant-Khachatrian

README.md

@@ -1,4 +1,16 @@
-R-NET implementation in Keras.
-Original paper: https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/r-net.pdf
-#### Implementation of the network described [here](https://www.microsoft.com/en-us/research/publication/mrc/).
-##### It is designed to work on the [SQuAD](https://rajpurkar.github.io/SQuAD-explorer/) dataset.
+# R-NET implementation in Keras
+
+This repository is an attempt to reproduce the results presented in the [technical report by Microsoft Research Asia](https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/r-net.pdf). The report describes a complex neural network called [R-NET](https://www.microsoft.com/en-us/research/publication/mrc/) designed for question answering.
+
+R-NET is currently (July 2017) the best model on Stanford QA database: [SQuAD](https://rajpurkar.github.io/SQuAD-explorer/). SQuAD dataset uses two performance metrics, exact match (EM) and F1-score (F1). Human performance is estimated to be EM=82.3% and F1=91.2%. 
+
+The report describes two versions of R-NET:
+1. One is marked as `R-NET (Wang et al., 2017)` (which refers to a paper which not yet available online) and reaches EM=71.3% and F1=79.7%. It consists of input encoders, a modified version of Match-LSTM, self-matching attention layer (the main contribution of the paper) and a pointer network. 
+2. The second version `R-NET (March 2017)` has one additional BiGRU between the self-matching attention layer and the pointer network and reaches EM=72.3% and F1=80.7%.
+
+The current best single-model on SQuAD leaderboard has a higher score, which means R-NET development continued after March 2017. Ensemble models reach higher scores.
+
+This repository contains an implementation of the first version, but we cannot yet reproduce the reported results. The best performance we got so far was EM=54.21% and F1=65.26%. We are aware of two differences between our implementation and the network described in the paper:
+
+1. We do not use character-level embedding at the input.
+2. The first formula in (11) of the [report](https://www.microsoft.com/en-us/research/wp-content/uploads/2017/05/r-net.pdf) contains a strange summand W_v^Q V_r^Q. Both tensors are trainable and are not used anywhere else in the network. We have replaced this product with a single trainable vector.