7 ms·

I'm sorry, but this is a really naive statement. They literally teach Page Rank in basic Linear Algebra classes. Does that mean a CS undergrad can recreate Goog

I'm sorry, but this is a really naive statement. They literally teach Page Rank in basic Linear Algebra classes. Does that mean a CS undergrad can recreate Google? Operating a service with real-time requirements at scale is a completely different beast than writing a proof-of-concept prototype.

Hydraulix989 7y ago
Understanding the concept is a very different thing than inventing it. When Google was built, Page Rank wasn't taught ANYWHERE because nobody thought of modeling web page relevance using a Markov chain yet (well, HITS came pretty close, but you get my point).
- jackcosgrove 7y ago
  PubMed was using a PageRank type algorithm before Google.
- Hydraulix989 7y ago
  @jackcosgrove: Source? As I understand, the closest prior art to PageRank was HITS (as I called out earlier).
  - jackcosgrove 7y ago
    I'm sorry, I was mistaken. It was Web of Science, not PubMed. Garfield, Garfield, Eugene. Citation indexing: Its theory and application in science, technology, and humanities. New York: Wiley, 1979, P. 1. http://garfield.library.upenn.edu/ci/chapter1.PDF http://garfield.library.upenn.edu/ci/chapter1.PDF During my investigation I found this article which claims an even earlier origin for this algorithm: https://m.phys.org/news/2010-02-google-pagerank-like-algorithm-dates.html https://m.phys.org/news/2010-02-google-pagerank-like-algorit...
    - Hydraulix989 7y ago
      These aren’t algorithms, they are just hand waving intuition that is “on the right track” but not quite there. You see similar “prior art” to Newtonian physics, for example, but there is still a huge leap from “oh, there might be something in these linkages” to “if we compute the stationary distribution of the Markov chain over the graph, we know the relative importance of each node.” Kleinberg was the first to algorithmically and rigorously codify this intuition.
Kye 7y ago
My understanding is that Google hasn't relied on Page Rank in a very long time. Depending on who you ask, they use 100-300+ factors to rank pages. Or they switched to an artificial brain that uses its own magic that no one really understands. I haven't kept up. All I know is Google is useless for all but the most obvious queries now.
- Hydraulix989 7y ago
  PageRank, although deceptively simplistic now (much like the concept of gravity), was much more difficult to invent than to understand. Starting your own Google 20 years ago would have been out of reach of your everyday CS student at the time, even though PageRank is the only thing you would have to “invent” per se (and even though today it is simple enough of a concept to be taught in undergrad). Today, starting a Google is even harder because of Google’s moat and the advancements such as the deep learning infrastructure used for ranking pages (which still can be replicated if you throw a few Stanford grads at the problem — they don’t even have to be budding Larry Pages in aptitude — but a working replica isn’t nearly enough to compete with Google’s foothold in the market).

Hydraulix989 7y ago

Understanding the concept is a very different thing than inventing it. When Google was built, Page Rank wasn't taught ANYWHERE because nobody thought of modeling web page relevance using a Markov chain yet (well, HITS came pretty close, but you get my point).

jackcosgrove 7y ago
PubMed was using a PageRank type algorithm before Google.
Hydraulix989 7y ago
@jackcosgrove: Source? As I understand, the closest prior art to PageRank was HITS (as I called out earlier).
- jackcosgrove 7y ago
  I'm sorry, I was mistaken. It was Web of Science, not PubMed. Garfield, Garfield, Eugene. Citation indexing: Its theory and application in science, technology, and humanities. New York: Wiley, 1979, P. 1. http://garfield.library.upenn.edu/ci/chapter1.PDF http://garfield.library.upenn.edu/ci/chapter1.PDF During my investigation I found this article which claims an even earlier origin for this algorithm: https://m.phys.org/news/2010-02-google-pagerank-like-algorithm-dates.html https://m.phys.org/news/2010-02-google-pagerank-like-algorit...
  - Hydraulix989 7y ago
    These aren’t algorithms, they are just hand waving intuition that is “on the right track” but not quite there. You see similar “prior art” to Newtonian physics, for example, but there is still a huge leap from “oh, there might be something in these linkages” to “if we compute the stationary distribution of the Markov chain over the graph, we know the relative importance of each node.” Kleinberg was the first to algorithmically and rigorously codify this intuition.

Kye 7y ago

My understanding is that Google hasn't relied on Page Rank in a very long time. Depending on who you ask, they use 100-300+ factors to rank pages. Or they switched to an artificial brain that uses its own magic that no one really understands. I haven't kept up. All I know is Google is useless for all but the most obvious queries now.

Hydraulix989 7y ago
PageRank, although deceptively simplistic now (much like the concept of gravity), was much more difficult to invent than to understand. Starting your own Google 20 years ago would have been out of reach of your everyday CS student at the time, even though PageRank is the only thing you would have to “invent” per se (and even though today it is simple enough of a concept to be taught in undergrad). Today, starting a Google is even harder because of Google’s moat and the advancements such as the deep learning infrastructure used for ranking pages (which still can be replicated if you throw a few Stanford grads at the problem — they don’t even have to be budding Larry Pages in aptitude — but a working replica isn’t nearly enough to compete with Google’s foothold in the market).