This blog will primarily be about reviewing, and, if possible, implementing interesting ideas from computer science.
The discipline of computer science is advancing at a galloping pace. On New Year’s day, I came across Adrian Colyer’s blog, The Morning Paper. Adrian has been reading and reviewing one interesting/influential CS paper every weekend for over a year. I have personally experienced and heard from others the difficulty of learning a new subject from papers. Textbooks are the best way to learn a subject. However, for many frontier topics in CS (and elsewhere), textbooks are hard to come by (is there a good text for Deep Reinforcement learning, as of February 2017?). I am going to take help from survey papers, PhD theses, and course material (where pedagogical insight has gone into the course structure).
Between the time I started writing this post and finished it, I saw a talk by Sean Cribbs. It was titled The Refeshingly Rewarding Realm of Research Papers. It gave me some good pointers: three passes through a new paper. The first pass involves a careful look at title-abstract-introduction, looking at section headings, conclusions, and ticking off references. The second pass involves reading more closely, looking at the proofs, graphs, and figures, and marking off unread references. The third pass involves stepping through closely, challenging all assumptions, and comparing thought processes. Sean suggests the following three tasks at the end:
- Summarize the paper in 1-2 sentences
- Outline major contributions
- Note strengths/weaknesses
Skimming through papers is something that I have no trouble doing. I have trouble with engaging in thought processes. Creating this blog is an attempt to making such things easier to do.
I also think that with Markdown syntax, IPython notebooks, and visualization tools (I am recalling Chris Olah, Michael Nielsen, and Andrej Karpathy’s usage of interactive visualizations), it is straight-forward to bring code into your engagement with peers and audience.
Let us see how this experiment will unfold.