4 ms·

"At Amazon we will be investing an order of magnitude more resources towards this problem. With data and computers to match this. This is significant leverage.

by glup 10y ago

"At Amazon we will be investing an order of magnitude more resources towards this problem. With data and computers to match this. This is significant leverage. Hence the change." The research questions will be in the interest of Amazon, which may not align with the public good (though, admittedly, the logic of the university doesn't always align with the public good). He'll get more leverage, but I'd be surprised if he works on the same problems. How about updating the academic funding model before everyone goes to industry?

eva1984 10y ago
In academia most professors are doing some or most of their projects for the source of funding, which I think is not too much about public good anyway. I think it is about time for machine learning researchers to go to industry, there isn't that much data in academia. Good public datasets are rare, like ImageNet, and have been basically overfitted to the core. In industry, it is totally another story.
- lqdc13 10y ago
  They are usually private datasets and details of the training are not usually disclosed. There could also be duplicated effort by every company competing in the field. There is also the side that it would make it harder for startups to compete as the progress is made on training on large datasets using huge clusters of machines. Such datasets and compute power is impossible to acquire for small companies.
- adrianN 10y ago
  So maybe we need to start funding the acquisition of large public datasets in academia?
  - newjersey 10y ago
    The fact that Stanford University creates patents from publicly funded research then turns around and sells them to non manufacturing entities makes me wary of pumping any new public money into research at universities. They're clearly corrupt to the core now.
    - thansharp 10y ago
      I haven't heard of this before. This is pretty interesting. Do you have a source?
      - rhinoceraptor 10y ago
        There was recently an episode of Planet Money about one instance of it: http://www.npr.org/sections/money/2016/06/10/481597112/episode-705-the-muscle-patents http://www.npr.org/sections/money/2016/06/10/481597112/episo...
  - taliar 10y ago
    Or somehow incentivize companies to follow an academic model.
jhbadger 10y ago
This sort of thing goes in cycles though. In the late 1990s when I was a postdoc, it seemed everyone in academia who could program or even just knew HTML was running off to a dot.com. A couple of years later most of them came crawling back on their knees begging for their old jobs.
autokad 10y ago
its not just about the data. academia doesnt pay the other workers well and have high turnover. If your a professor trying to run a research lab, your stellar employees wont stay for very long and you'll have high turnover.
taliar 10y ago
There's a trade-off here: You have to do things that have immediate value to the company (which, at Amazon, corresponds exactly to things that have value to your customers), but you pretty much have the power of the entire internet to do this (AWS). Not to mention more data than most researchers could ever dream of. It's not just monetary. Since research at Amazon is new, I have no idea how well they're going to handle giving back to the community long-term. But I'm interning with a verification group there and the gist seems to be that if the group wants to give back and there's at least one person who's influential enough to get what they ask for, the group will give back. I think it's actually problematic that research and industry concerns don't correspond more often. Yeah, it's not our job to be engineers and ship products, but it seems like there are important questions we ought to be answering as researchers that we consider industry ground, but that engineers don't actually have the background to solve. Some of these are totally arbitrary, chosen by the community, deemed "uninteresting" if you want to get published anywhere decent, even though there's plenty of interesting work left.

by glup 10y ago

eva1984 10y ago

In academia most professors are doing some or most of their projects for the source of funding, which I think is not too much about public good anyway. I think it is about time for machine learning researchers to go to industry, there isn't that much data in academia. Good public datasets are rare, like ImageNet, and have been basically overfitted to the core. In industry, it is totally another story.

lqdc13 10y ago
They are usually private datasets and details of the training are not usually disclosed. There could also be duplicated effort by every company competing in the field. There is also the side that it would make it harder for startups to compete as the progress is made on training on large datasets using huge clusters of machines. Such datasets and compute power is impossible to acquire for small companies.
adrianN 10y ago
So maybe we need to start funding the acquisition of large public datasets in academia?
- newjersey 10y ago
  The fact that Stanford University creates patents from publicly funded research then turns around and sells them to non manufacturing entities makes me wary of pumping any new public money into research at universities. They're clearly corrupt to the core now.
  - thansharp 10y ago
    I haven't heard of this before. This is pretty interesting. Do you have a source?
    - rhinoceraptor 10y ago
      There was recently an episode of Planet Money about one instance of it: http://www.npr.org/sections/money/2016/06/10/481597112/episode-705-the-muscle-patents http://www.npr.org/sections/money/2016/06/10/481597112/episo...
- taliar 10y ago
  Or somehow incentivize companies to follow an academic model.

jhbadger 10y ago

This sort of thing goes in cycles though. In the late 1990s when I was a postdoc, it seemed everyone in academia who could program or even just knew HTML was running off to a dot.com. A couple of years later most of them came crawling back on their knees begging for their old jobs.

autokad 10y ago

its not just about the data. academia doesnt pay the other workers well and have high turnover. If your a professor trying to run a research lab, your stellar employees wont stay for very long and you'll have high turnover.

taliar 10y ago

There's a trade-off here: You have to do things that have immediate value to the company (which, at Amazon, corresponds exactly to things that have value to your customers), but you pretty much have the power of the entire internet to do this (AWS). Not to mention more data than most researchers could ever dream of. It's not just monetary. Since research at Amazon is new, I have no idea how well they're going to handle giving back to the community long-term. But I'm interning with a verification group there and the gist seems to be that if the group wants to give back and there's at least one person who's influential enough to get what they ask for, the group will give back. I think it's actually problematic that research and industry concerns don't correspond more often. Yeah, it's not our job to be engineers and ship products, but it seems like there are important questions we ought to be answering as researchers that we consider industry ground, but that engineers don't actually have the background to solve. Some of these are totally arbitrary, chosen by the community, deemed "uninteresting" if you want to get published anywhere decent, even though there's plenty of interesting work left.