Deep Learning Papers Survey

Personal repository to track my paper surveys. Based on Yagami360/MachineLearning-Papers_Survey.

Each paper summary is under Issues.

• Papers I intend to start soon are labeled TODO.
• Started papers are labeled in progress. Some papers only need a cursory glance and will thus stay under this label.
• Nearly finished papers are labeled almost done. Most of my "finished" papers will end up here because I may want to add more in the future.
• 100% finished papers are under Closed.

I'll only list nearly or 100% finished papers in this document.

■ Deep Learning Papers

◎ General Deep Learning

• TODO

◎ Vision

• TODO

◎ Generative Graphics

↳ Semantic Image Synthesis

• [GauGAN/SPADE] Semantic Image Synthesis with Spatially-Adaptive Normalization - Park et al. 2019 CVPR
• [SMIS] Semantically Multi-modal Image Synthesis - Zhu et al. 2020

↳ Latent Space Exploration

• [SeFa] Closed-Form Factorization of Latent Semantics in GANs - Shen and Zhou 2020

↳ Metrics

• The Unreasonable Effectiveness of Deep Features as a Perceptual Metric - Zhang et al. 2018 CVPR

◎ Reinforcement Learning

• TODO

↳ Model Free

↳ Model Based

↳ Learning from Demonstrations

◎ Natural Language Processing

• TODO

■ Other Papers

• TODO: Maybe I'll survey papers from other fields.

■ How to read and research papers on machine learning

• Paper survey

  • The easiest way to find an advanced version of the paper is to find the cited paper on Google Scholar.
  • Basically, the ones with the highest number of citations are the ones of interest. However, the latest papers will of course have fewer citations.
  • Basically, the latest papers give better experimental results, so look for the latest published date on Google Scholar as much as possible.
  • If there is no official implementation of the paper, the cost of reproducing the original implementation is high, so we preferentially search for one with an official implementation.
  • Check PapersWithCode to see if the paper is officially implemented.

• How to read a dissertation

  • Basically, it is efficient to read "Abstract"-> "Introduction"-> "Conclusion"-> "Details of what you did"-> "Experiments"-> "Related Work". Related Work can be skipped at worst.
  • Notice the words "In this paper ..." and "In this work ..." in the Abstract and Introduction, and the words "the contributions of this paper ... as follows." At the end of the Introduction.
  • Unresolved issues and problems in the paper are often described in Future work and Conclusion in the paper. Or, it is often mentioned in the Related Work of the paper citing the paper.

■ Template for Paper Survey

You can use the GitHub issue template that I created for your own purposes.

■ Reference site

◎ Convenient sites

• PapersWithCode.com: check if a paper is implemented
• SemanticScholar.org: uses ML to cluster related papers, create custom paper feeds
• ConnectedPapers.com: visualize a graph of related papers

◎ Other reference sites

• Yagami360/MachineLearning-Papers_Survey
• ymym3412/acl-papers
• shunk031/paper-survey