[Adapt] Weekly Seminar

[洪韫妍]

Hi Adapters：
     For seminar today, I will introduce a model to optimize the word generator part in seq2seq model. It's proposed in the paper "Query and Output: Generating Words by Querying Distributed Word Representations for Paraphrase Generation"(https://arxiv.org/abs/1803.01465). In this paper, instead of using traditional linear transformation and softmax function to generate the words, the author learn from attention mechanism, and generates the words according to the distributed word representations in a retrieval style.

    I think the representation can give us inspiration to do more research on basic seq2seq model.

    See you then!
     
Time:17:00 Nov 21
Venue:SEIEE 3-517A
Best,
Sophia.