4 ms·

Reading the article and some of the links make me feel like the author Jürgen Schmidhuber is the academic version of the patent troll. It sounds like he publis

Reading the article and some of the links make me feel like the author Jürgen Schmidhuber is the academic version of the patent troll. It sounds like he published some theoretical musings back in 1990s without any real practical implementation that did anything useful and since then has run around accusing AI researchers who actually produced concrete research and techniques to get are actually in use today of plagiarism.

aborsy 3y ago
Regardless, it should be easy to verify. He cites a paper and says it’s done in this earlier paper. On the other hand, if his claims are verified, some of these research “discoveries” are simply common sense and would occur to most people working on the subject. HLB were awarded to a good extent because they worked on deep learning at the right time. Deep learning became hugely practical, and outperformed the state of the art in many applications. HL also worked for major companies.
logicchains 3y ago
>It sounds like he published some theoretical musings back in 1990s without any real practical implementation He published working models back then, the problem was compute power was very limited. In the past decade deep learning took off, people took his models, renamed then and ran them on vastly more powerful computers, to great success, then failed to cite him.
- vbarrielle 3y ago
  It's actually a common problem in lots of computing related research communities. Papers older than 10 years ago, sometimes even 5 years ago, are ignored. Because the results do not compare in terms of computing power, and because reimplementing old papers is often very hard because they are often too vague.
P-NP 3y ago
> without any real practical implementation Don't you know that billions of people are using his work on a daily basis on their smartphone? CV: https://people.idsia.ch/~juergen/cv.html https://people.idsia.ch/~juergen/cv.html

aborsy 3y ago

Regardless, it should be easy to verify. He cites a paper and says it’s done in this earlier paper. On the other hand, if his claims are verified, some of these research “discoveries” are simply common sense and would occur to most people working on the subject. HLB were awarded to a good extent because they worked on deep learning at the right time. Deep learning became hugely practical, and outperformed the state of the art in many applications. HL also worked for major companies.

logicchains 3y ago

>It sounds like he published some theoretical musings back in 1990s without any real practical implementation He published working models back then, the problem was compute power was very limited. In the past decade deep learning took off, people took his models, renamed then and ran them on vastly more powerful computers, to great success, then failed to cite him.

vbarrielle 3y ago
It's actually a common problem in lots of computing related research communities. Papers older than 10 years ago, sometimes even 5 years ago, are ignored. Because the results do not compare in terms of computing power, and because reimplementing old papers is often very hard because they are often too vague.

P-NP 3y ago

> without any real practical implementation Don't you know that billions of people are using his work on a daily basis on their smartphone? CV: https://people.idsia.ch/~juergen/cv.html https://people.idsia.ch/~juergen/cv.html