First, caveats- 1. The below tips apply to ML and bioeng, which are pretty different fields but I doubt are totally general 2. Writing down tacit knowledge is hard. 3. There must be many ways to do this- so share yours! Now for the ideas: (2/n)
-
-
Prikaži ovu nit
-
Grocking literature is ridiculously important. Depending on the project, it may be *most* of your job as a researcher. You can’t come up with something new without knowing what exists. And nobody teaches you how to do it, especially jumping into a new field(3/n)
Prikaži ovu nit -
Figure out external markers of quality. Most reputable journals, well respected conferences, high profile labs, successful commercializations, citations etc. Don’t take these as gospel, ask why they are the way they are (actually high quality? funder priorities? media hype?) 4/n
Prikaži ovu nit -
Be direct. Have a goal for what you want to know. Search for keywords as close as you can to this thing. If nothing lands, think about what source (eg review paper or blog post) would intro the jargon. Find the highest “external” quality papers in these results. 5/n
Prikaži ovu nit -
Search breadth first. From initial hits, skim intros, discussions and citation list. Collect citations like Pokémon. Pay attention to *how* citations are made “in the seminal work of” vs laundry list(12,23...). Evaluate “external quality” and location in citation network. 6/n
Prikaži ovu nit -
Focus on high leverage papers. Once you have a picture of the citation network, pick the paper(s) which looks highest value of info (usually on the quality/network centrality/ proximity to your core question frontier). Some papers are *much* more useful than others. 7/n
Prikaži ovu nit -
They are lying to you. Hopefully not deliberately, but ~every paper has errors which will someday be overturned. Don’t believe anything, especially in high external quality papers. Make them convince you from first principles. 8/n
Prikaži ovu nit -
Work with uncertainty. If you don’t have the background assumed by a paper, don’t give up or decide to find an intro course online. Read the whole thing. Note what you don’t understand, and consider all the plausible explanations for what is going on, given what you know. 9/n
Prikaži ovu nit -
Prioritize ruthlessly. Given what you identified as potential errors/issues with the paper, and gaps in your knowledge, scan the citation networks again with an eye to minimize your new uncertainty. Google scholar
@SemanticScholar fwd and backwd citations are your friend. 10/nPrikaži ovu nit -
Look for negative space. While you look for open problems and established techniques, remember: there is a *strong* selection filter. Much more is below the surface. Why is this obvious technique not used? Why so few citations? How many neg. results before this one? 11/n
Prikaži ovu nit -
Train your intuition. As you read, make predictions about what you expect to find elsewhere in the literature. How will papers that cite this one improve on it? When you pattern match a paper as useful to read, is it actually? Check your guesses. 12/n
Prikaži ovu nit -
And talk to people! Someone who is clued in can occasionally show you right to the-paper-you-wish-you-read. I’ll stop there for now! There is much more to say and always more to learn. What is your approach to literature? 13/13
Prikaži ovu nit
Kraj razgovora
Novi razgovor -
Čini se da učitavanje traje već neko vrijeme.
Twitter je možda preopterećen ili ima kratkotrajnih poteškoća u radu. Pokušajte ponovno ili potražite dodatne informacije u odjeljku Status Twittera.