如何快速了解新研究领域及选择研究方向

点击查看【bilibili】

核心内容:
1.快速适应新研究领域/主题
2.找到潜在研究方向

Agenda

  • Information acquisition 信息获取
  • Paper reading 论文阅读
    • How many papers 多少论文
    • How to ？怎么读
    • What extend ？你的期望
  • Research direction selection and evaluation 研究方向选择和演化

Information acquisition

  • Survey paper 综述
  • Textbook （undergrad introductory courses）本科入门课程
  • YouTube / Bilibili （科普视频快但也会有很多错误）
  • Course videos
    • Coursera
    • MIT Open Courseware
    • MOOC
    • Other online course materials

Paper reading

文献是不可能读完的，那么读到什么程度呢？

How many ？

  • By numbers：【通过数量】
    • 5-20：brief idea 吴恩达老师提到过5-20基本可以有一些快速的idea
    • 30-40：basic understanding（基础认知） 个人经验 （在一些方向上反复横跳）
    • 50-100：understanding of the domain（对整个领域有了认知）
  • OR
  • By selection：【通过选择】
    • Recent work in 5 years (or 1-3 years, if it’s a super fast developing area) 近5年的工作
    • Paper with high citation number, or the“fundamental work” 高引用的文章/开山之作

image.png
好玩最重要！

How to ？

3-round method：“How to read a paper? 怎样阅读一篇论文？”by Keshav，S 三遍法

  • First round: title, abstract, intro, headings, conclusion, reference (glance, can

ignore right now)

  • Answer: What Type, Context(relations to others?), Contribution?
  • Second round: Content
  • Answer: what was this paper doing?
  • Third round: re-implementation (can ignore if not doing that)

Andrew Ng’s Progress check：

  • 10-20%: Title, Abstract, Number(Conclusion)
  • 20-40%: Intro, Conclusion, Number, Technical details (Experiment)
  • 40-60%: Algorithm
  • 60-80%: Mathematical Details, Implementation
  • 80-100%: Q: What does the authors tried to do? What is the key point? If |

were the author what will I do?

My own method(adapt from Professor Springer’s course note):

  • Issue?
  • Current work done?
  • Gap?
  • How to bridge the gap?
  • What are the mainstream methods available?

What extend？

  • Depends on your need 看你需求
    • LR or survey writing 写综述 100起步
      • 100+ papers at least, depends on the area
      • OR
      • Fundamental work(origin), highly cited work, 3-5 years most recent work
  • Understanding of the area 了解领域 50-100 够了 / 综述 + 开山之作 + 高引用5-10
    • 50-100 will do according to Dr. Andrew Ng
    • OR
    • Surveys + fundamental work + most highly cited (top 5- top 10?)
  • Decide research direction? 研究方向
    • Next page

Research direction decision

  • YOU make the decision since YOU do the research work 首先，是你做研究，先要感兴趣
  • Self evaluation
    • What skills/knowledge are required? 这个领域需要哪些技能？
    • What skills/knowledge | have acquired? 你有那些技能？
    • How long will it take to learn the new required skills/knowledge? 差的那些技能你大概还要多久才能学会
    • What facilities are available? 学校能提供那些设备
    • If not available is there any way to overcome it? 如果没有能否克服
    • What skills can be transferred? 另外就是你现有的技能能够有多少转化
    • MATHS!!! 数学
    • STATISTIC!! 统计学

参考

  • Keshav, S. (2007). How to read a paper. ACM SIGCOMM Computer

Communication Review, 37(3), 83-84.

  • John A. Springer, CNIT 58100, Course notes, Fall 2016 and Fall 2017
  • Stanford CS230: Deep Learning| Autumn 2018| Lecture 8 - Career

Advice / Reading Research Papers