Dynamic Memory Networks for Question Answering over Text and Images

学习做一个工程，，

In order to achieve this goal, we face two major obstacles. Many

NLP tasks use different architectures, such as TreeLSTM (Tai et al.,

2015) for sentiment analysis, Memory Network (Weston et al., 2015)

for question answering, and Bi-directional LSTM-CRF (Huang et al.,

2015) for part-of-speech tagging. The second problem is full multitask

learning tends to be very difficult, and transfer-learning remains

to be a major obstacle for current neural network architectures across

artificial intelligence domains (computer vision, reinforcement learning,

etc.).