Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Geoffrey Hinton has joined Google

plus.google.com

Post reply on HN
Print view View original

91–100 of 102 posts

Re: Geoffrey Hinton has joined Google

#91

Post by xyzzy123 »

kolektiv wrote: post #3

Sigh. I don't know whether I'm alone in being saddened that many of the excellent thinkers of our time, who once might have spent their lives probing the the very limits of our comprehension in their field, now end up comfortably turning their minds to the services of (in the long term) rather short sighted commercial ends. I can imagine that this sentiment is probably against the prevailing HN mood, but I've been th…

What if you consider the application a trojan horse for doing interesting things which are useful well outside the supposed domain of application?

I think the trope of "the best minds working on ad clicks" is perhaps not giving people enough credit.

Hackers and academics always find a way to screw around and work on interesting things. The trick is to find interesting structure in mundane problems.

Re: Geoffrey Hinton has joined Google

#93

Post by mindcrime »

kolektiv wrote: post #3

Sigh. I don't know whether I'm alone in being saddened that many of the excellent thinkers of our time, who once might have spent their lives probing the the very limits of our comprehension in their field, now end up comfortably turning their minds to the services of (in the long term) rather short sighted commercial ends. I can imagine that this sentiment is probably against the prevailing HN mood, but I've been th…

The point of a university used to be being the highest pinnacle of thought. A place where people could have everything not related to intellectual pursuit taken care of so they could devote themselves to the pursuit of greater knowledge.

Sounds good on paper, but I'd question if that was ever actually true in reality anyway.

Now that seems to be replaced by a corporate campus

And? Is research less valuable because it's done by a corporation? Look at how much basic research IBM sponsors and how much Bell Labs and others have sponsored in the past. If anything, I would go the opposite direction and lament the decline in corporate spending on fundamental research. I'd like to see more academics go into the business world.

"Startups are the new graduate school" - can't remember where I heard that quote, but I think there's something to it.

Re: Geoffrey Hinton has joined Google

#94

Post by slightperturb »

stiff wrote: post #75
kolektiv wrote: post #3

Sigh. I don't know whether I'm alone in being saddened that many of the excellent thinkers of our time, who once might have spent their lives probing the the very limits of our comprehension in their field, now end up comfortably turning their minds to the services of (in the long term) rather short sighted commercial ends. I can imagine that this sentiment is probably against the prevailing HN mood, but I've been th…

It remains to be proved that an university is particularly better at innovation that anything else. At least in Computer Science most major historic breakthroughs happened in army departments and in commercial laboratories (Bell Labs, Xerox Parc) and today they are still strong, for example many of the most cited CS publications come from Microsoft Research. Google faces some of the biggest challenges in Machine Lear…

Of the most-cited 20 papers in CS (from citeseer), the affiliation of the first author of all but four is with academia:

Academia --------

1) Maximum likelihood from incomplete data via the EM algorithm, 1977. Dempster, Harvard.

2) Communicating sequential processes, 1978. Hoare, Oxford at time of publication.

4) Chord: A scalable peer-to-peer lookup service for internet applications, 2001. Stocia, Berkeley.

5) Distinctive Image Features from Scale-Invariant Keypoints, 2004. Lowe, UBC.

6) Induction of Decision Trees, 1986. Quinlan, New South Wales Institute of Technology.

7) Reinforcement Learning: An introduction, 1998. Sutton, UAlberta.

9) Graph-based Algorithms for Boolean Function Manipulation, 1986. Bryant, CalTech.

10) Scheduling Algorithms for Multiprogramming in a Hard-Real-Time Environment, 1973. Liu, MIT.

11) The anatomy of a large-scale hypertextual web search engine, 1998. Brin, Stanford.

12) A method for obtaining digital signatures and public-key cryptosystems, 1978. Rivest, MIT.

14) A scalable content-addressable network, 2001. Ratnasamy, Berkeley.

15) New directions in cryptography, 1976. Diffie, Stanford.

16) Eigenfaces for recognition, 1991. Turk, MIT.

17) Authoritative sources in a hyperlinked environment, 1999. Kleinberg, Cornell.

18) Indexing by latent semantic analysis, 1990. Deerwester, UChicago.

20) Handbook of Applied Cryptography, 1996. Menezes, UWaterloo.

Industry --------

3) A tutorial on hidden Markov models and selected applications in speech recognition, 1989. LR Rabiner, Bell Labs.

8) Optimization by simulation annealing, 1983. Kirkpatrick, IBM.

13) Snakes — active contour models, 1987. Kass, Schlumberger Palo Alto Research.

19) Fast algorithms for mining association rules, 1994. Agrawal, IBM.

Re: Geoffrey Hinton has joined Google

#96

Post by jhartmann »

boucher wrote: post #34
Locke1689 wrote: post #24

Earlier quoted context omitted.

To be honest... kind of. They release papers but most of the layers at the bottom are Google proprietary. Certainly the source code is proprietary. Hell, even most of the information about GFS is proprietary, despite having a successor. From an academic perspective, their disclosure is kind of crap.

It's always been interesting to me that someone publishing a paper in computer science (especially someone working at a University) would not want to open source the code that produced those results. It's the best kind of repeatable experiment. I recently went looking for any code published by Hinton, since I've been doing some neural network research as a hobby, but was unable to find any at all.

Actually, Alex Krizhevsky published some of the code they have been using @ http://code.google.com/p/cuda-convnet/

It seems from my experience that generally Hinton and team seem very open to sharing quite abit of their code, data and results.

Re: Geoffrey Hinton has joined Google

#97

Post by stiff »

slightperturb wrote: post #94
stiff wrote: post #75

Earlier quoted context omitted.

It remains to be proved that an university is particularly better at innovation that anything else. At least in Computer Science most major historic breakthroughs happened in army departments and in commercial laboratories (Bell Labs, Xerox Parc) and today they are still strong, for example many of the most cited CS publications come from Microsoft Research. Google faces some of the biggest challenges in Machine Lear…

Of the most-cited 20 papers in CS (from citeseer), the affiliation of the first author of all but four is with academia: Academia -------- 1) Maximum likelihood from incomplete data via the EM algorithm, 1977. Dempster, Harvard. 2) Communicating sequential processes, 1978. Hoare, Oxford at time of publication. 4) Chord: A scalable peer-to-peer lookup service for internet applications, 2001. Stocia, Berkeley. 5) Disti…

This includes textbooks ("Handbook of Applied Cryptography", "Reinforcement learning: an introduction") and introductory articles ("A tutorial on hidden Markov models"). Also, ironically, paper 11 in "Academia" is about Google.

Anyway, there are strong incentives at the university to produce papers, to the point of this being one of the top priorities for every university worker. So, I don't wonder the universities are the best at producing papers that are cited, I just wanted to emphasize that even in that area the industry has not bad contributions. There is also lots of innovation happening in the industry that is not published in the form of papers, since in many places there are no incentives to write up one after doing something of value.

Re: Geoffrey Hinton has joined Google

#98

Post by slightperturb »

stiff wrote: post #97
slightperturb wrote: post #94

Earlier quoted context omitted.

Of the most-cited 20 papers in CS (from citeseer), the affiliation of the first author of all but four is with academia: Academia -------- 1) Maximum likelihood from incomplete data via the EM algorithm, 1977. Dempster, Harvard. 2) Communicating sequential processes, 1978. Hoare, Oxford at time of publication. 4) Chord: A scalable peer-to-peer lookup service for internet applications, 2001. Stocia, Berkeley. 5) Disti…

This includes textbooks ("Handbook of Applied Cryptography", "Reinforcement learning: an introduction") and introductory articles ("A tutorial on hidden Markov models"). Also, ironically, paper 11 in "Academia" is about Google. Anyway, there are strong incentives at the university to produce papers, to the point of this being one of the top priorities for every university worker. So, I don't wonder the universities a…

I think you should reconsider your take on that paper 11. It's not "about" Google-- it is the cornerstone work that led to the creation of Google-- and it was written by two academics (PhD candidates at Stanford). In your previous post, you mentioned "the somewhat vacuous atmosphere of an universities[sic] thinking". But you need to step back and realize that this paper is showing that the university environment led to the creation of the key ideas behind Google, and thus I don't see justification to call it a "vacuous atmosphere".

In fact, I believe what Google is doing so well (and what is attracting great academic researchers like Hinton) is the creation of the rich and rewarding atmosphere mimicking that of classic academia. (And at the same time, cuts to education and research budgets are helping to erode the benefits of a life in the academy.)

Re: Geoffrey Hinton has joined Google

#99

Post by ok_craig »

OGinparadise wrote: post #82
ok_craig wrote: post #44

Earlier quoted context omitted.

The common lament about Google being just a place that optimizes ad clicks is overused, and lazy thinking. I don't even think it's a conclusion that's the result of actual thought anymore, just reverb. Google is a big company. It would not even be worth it or effective to have every single engineer dedicated to this ad click business. There's a team for that, and there are lots of other teams, for lots of other inter…

The common lament about Google being just a place that optimizes ad clicks is overused, and lazy thinking....many of them do serious intellectual research Now that's lazy thinking. No one accused Google or Verizon or of not having "many people" that do serious research. Personally, I see absolutely no problem with what he did, he has a family and his future to worry about. After all these years he deserves to make a…

> Now that's lazy thinking. No one accused Google or Verizon or of not having "many people" that do serious research.

By assuming that every single person who joins Google is somehow being used to optimize ad clicks, the implication is that they are abandoning whatever serious intellectual research they could have otherwise been doing.

> There's no trick in the book they do not use to make you click on ads or sign for their services.

This doesn't counter my point. The fact that their ad click team is actually doing its job doesn't mean anything about the greater motivation of the company.

Re: Geoffrey Hinton has joined Google

#100

Post by stiff »

slightperturb wrote: post #98
stiff wrote: post #97

Earlier quoted context omitted.

This includes textbooks ("Handbook of Applied Cryptography", "Reinforcement learning: an introduction") and introductory articles ("A tutorial on hidden Markov models"). Also, ironically, paper 11 in "Academia" is about Google. Anyway, there are strong incentives at the university to produce papers, to the point of this being one of the top priorities for every university worker. So, I don't wonder the universities a…

I think you should reconsider your take on that paper 11. It's not "about" Google-- it is the cornerstone work that led to the creation of Google-- and it was written by two academics (PhD candidates at Stanford). In your previous post, you mentioned "the somewhat vacuous atmosphere of an universities[sic] thinking". But you need to step back and realize that this paper is showing that the university environment led…

Please don't have the impression that I am being condescending toward universities, I respect and admire a lot of work that is being done there, I just think that not all of innovation can happen there. Google uses a lot of theory, but they also act in the real world and face real-world problems that can inspire new research and new theories, whereas at an university you are to a large extent cut off from such stimulus.
Post reply on HN