[Adapt] [Seminar] Introduction to REINFORCE Algorithm in Neural Sequence Generation

[racheter at sjtu.edu.cn]

Hi Adapters,

In today's weekly seminar, I will give an introduction to the application of REINFORCE algorithm in some sequence (mainly text) generation tasks. REINFORCE is a traditional reinforcement learning algorithm, and some recent papers apply it on sequence generation tasks like summarization and image captioning to tackle some problems of maximum likelihood estimation and to directly optimize the evaluation metric.

Here are the papers I’ll briefly talk about:
1. Sequence Level Training with Recurrent Neural Networks: https://arxiv.org/abs/1511.06732
2. Self-critical Sequence Training for Image Captioning: https://arxiv.org/abs/1612.00563
3. A Deep Reinforced Model for Abstractive Summarization: https://arxiv.org/abs/1705.04304

Time: June 6th, Wed 17:00
Venue: Room 3-517

See you then!

Best,
Yifan