Difference between revisions of "Fall 2017 CS595I Advanced NLP/ML Seminar"

From courses
Jump to: navigation, search
 
(7 intermediate revisions by the same user not shown)
Line 4: Line 4:
 
If you registered this class, you should contact the instructor to lead the discussion of one paper below.
 
If you registered this class, you should contact the instructor to lead the discussion of one paper below.
 
If you don't lead the discussion, you will then need to write a 3-page final report in NIPS 2017 style,  
 
If you don't lead the discussion, you will then need to write a 3-page final report in NIPS 2017 style,  
comparing any two of the papers below.
+
comparing any two of the papers below. '''Due: 12/18, 23:59pm PT''' to william@cs.ucsb.edu.
  
 
*09/26:  
 
*09/26:  
Line 36: Line 36:
 
*11/14:
 
*11/14:
 
** Richard: Understanding Black-box Predictions via Influence Functions, Koh and Liang, ICML 2017 Best Paper. https://arxiv.org/pdf/1703.04730.pdf
 
** Richard: Understanding Black-box Predictions via Influence Functions, Koh and Liang, ICML 2017 Best Paper. https://arxiv.org/pdf/1703.04730.pdf
 +
** Ke: Deep reinforcement learning from human preferences, Christiano et al., https://arxiv.org/pdf/1706.03741.pdf
  
 
*11/21:
 
*11/21:
 
** Conner: Attention Is All You Need, Vaswani et al., 2017 https://arxiv.org/abs/1706.03762
 
** Conner: Attention Is All You Need, Vaswani et al., 2017 https://arxiv.org/abs/1706.03762
 +
** Xiaoyong: Adversarial Feature Matching for Text Generation, Zhang et al., ICML 2017 http://proceedings.mlr.press/v70/zhang17b/zhang17b.pdf
  
 
*11/28:
 
*11/28:
 +
** Austin: Deep Reinforcement Learning that Matters, Henderson et al., arxiv https://arxiv.org/pdf/1709.06560.pdf
 +
** Adam: Generalization in Deep Learning, https://arxiv.org/pdf/1710.05468.pdf
  
 
*12/05: No meeting, NIPS conference.
 
*12/05: No meeting, NIPS conference.
Line 54: Line 58:
 
===Reinforcement Learning===
 
===Reinforcement Learning===
 
* Modular Multitask Reinforcement Learning with Policy Sketches, Andreas et al., ICML 2017 https://arxiv.org/pdf/1611.01796.pdf
 
* Modular Multitask Reinforcement Learning with Policy Sketches, Andreas et al., ICML 2017 https://arxiv.org/pdf/1611.01796.pdf
* Deep reinforcement learning from human preferences, Christiano et al., https://arxiv.org/pdf/1706.03741.pdf
 
* Deep Reinforcement Learning that Matters, Henderson et al., arxiv https://arxiv.org/pdf/1709.06560.pdf
 
 
* Robust Imitation of Diverse Behaviors, Wang et al. 2017, https://arxiv.org/pdf/1707.02747.pdf
 
* Robust Imitation of Diverse Behaviors, Wang et al. 2017, https://arxiv.org/pdf/1707.02747.pdf
 
* Programmable Agents, Denil et al., https://arxiv.org/pdf/1706.06383v1.pdf
 
* Programmable Agents, Denil et al., https://arxiv.org/pdf/1706.06383v1.pdf
Line 67: Line 69:
 
===Generation===
 
===Generation===
 
* Generating Sentences by Editing Prototypes, Guu et al., arxiv https://arxiv.org/abs/1709.08878
 
* Generating Sentences by Editing Prototypes, Guu et al., arxiv https://arxiv.org/abs/1709.08878
* Adversarial Feature Matching for Text Generation, Zhang et al., ICML 2017 http://proceedings.mlr.press/v70/zhang17b/zhang17b.pdf
 
 
* Adversarially Regularized Autoehttp://www.aclweb.org/anthology/D17-1120ncoders for Generating Discrete Structures, Zhao et al., https://arxiv.org/pdf/1706.04223.pdf
 
* Adversarially Regularized Autoehttp://www.aclweb.org/anthology/D17-1120ncoders for Generating Discrete Structures, Zhao et al., https://arxiv.org/pdf/1706.04223.pdf
  

Latest revision as of 17:00, 28 November 2017

Time: Tuesday 5-6pm. Location: HFH 1132.

If you registered this class, you should contact the instructor to lead the discussion of one paper below. If you don't lead the discussion, you will then need to write a 3-page final report in NIPS 2017 style, comparing any two of the papers below. Due: 12/18, 23:59pm PT to william@cs.ucsb.edu.

  • 09/26:
    • Mahnaz Summer research presentation: Reinforced Pointer-Generator Network for Abstractive Summarization.
    • Xin: FeUdal Networks for Hierarchical Reinforcement Learning, Vezhnevets et al., ICML 2017 https://arxiv.org/pdf/1703.01161.pdf
  • 12/05: No meeting, NIPS conference.
  • 12/12: No meeting, NAACL deadline.

Word Embeddings

Relational Learning and Reasoning

Reinforcement Learning

Generation

Dialog

NLP for Computational Social Science