It's news because that's the cold place neural networks went to wait out the "AI winter".
It's amazing that neural networks actually work now. They've been around since the 1960s, and for four decades, they sucked. They're only a little smarter; it's mostly sheer compute power applied to algorithms that fit on one screen of Matlab.
Highly glorifying, badly credited article that reads like a state propaganda piece. Hinton's work is amazing and his contribution has been immense, but this article is a joke.
Such science "journalism" is seriously hurting the field, including the researchers that are lionized in these articles. It also makes large groups of researchers who've been working hard on advancing the field think they don't matter in the eyes of the public, and that can be quite demoralizing.
That "unsupervised cats" paper from 2012 had nothing new in it at the time, although it was a feat of parallelization on a cluster of CPUs. Three years later its (uninteresting) results are not used anywhere. But Google used it as a PR piece at the time, and hundreds of journalists have been presenting it as some kind of game-changing breakthrough. It continues to this day.
Keep in mind that this is the Toronto Star. A local paper here in Toronto for obviously a broad audience outside of tech.
For a country and city that has a heavy inferiority complex and suffers greatly from the "Brain Drain" of technical talent to the US, this is one of those articles that's meant to inspire. Not an article for Canadians/Torontonians to puff their chests.
So I wouldn't be so critical about it's so called "propaganda" aspect.
I'm surprised about all of the backlash on here against the article. Look at the ImageNet competition - when it first started out the algorithms being used were very different hand-designed algorithms, then all of a sudden backprop dominated.
If you're interested in this area, my company is designing hardware accelerators for these algorithms. Our versions of the algorithms have a few extra twists to them to account for the analog nature of the hardware. I'm probably a bit biased, but I think this is a really interesting place to be working right now
Below is what posted on this month's Who is Hiring:
Isocline - Austin, TX - Software Engineer for High Performance Computing and Modeling
We are looking for two people - one interested in neural networks and one interested in GPS.
We are developing microchips that yield a 10-1000x improvement in performance & energy-efficiency compared to digital ASICs, GPUs, and FPGAs. We are a bootstrapped company and are fully funded through mid 2016. Patents pending.
Incredible. It is exciting to see which areas of life AI will optimize in the future. The medical company which is focusing on teaching AI to interpret patient charts and etc, is an excellent example. I wonder if it would be possible to build a deep AI for interpreting law, effectively replacing research work which lawyers do. It seems law has a methodical structure that AI could learn to navigate recognizing past precedents, performing wide cross referencing, exploiting existing loop holes in the system, and validating judicial decisions.
12 comments
[ 0.22 ms ] story [ 34.1 ms ] threadIt's amazing that neural networks actually work now. They've been around since the 1960s, and for four decades, they sucked. They're only a little smarter; it's mostly sheer compute power applied to algorithms that fit on one screen of Matlab.
- proper initialization of weights (should have been obvious, but no one really saw it)
- rectified linear units and dropout
Such science "journalism" is seriously hurting the field, including the researchers that are lionized in these articles. It also makes large groups of researchers who've been working hard on advancing the field think they don't matter in the eyes of the public, and that can be quite demoralizing.
That "unsupervised cats" paper from 2012 had nothing new in it at the time, although it was a feat of parallelization on a cluster of CPUs. Three years later its (uninteresting) results are not used anywhere. But Google used it as a PR piece at the time, and hundreds of journalists have been presenting it as some kind of game-changing breakthrough. It continues to this day.
For a country and city that has a heavy inferiority complex and suffers greatly from the "Brain Drain" of technical talent to the US, this is one of those articles that's meant to inspire. Not an article for Canadians/Torontonians to puff their chests.
So I wouldn't be so critical about it's so called "propaganda" aspect.
Toronto Professor, contributes to computer science subfield AI.
Vs
How a Toronto Professor, changed the world with Deep Learning.
Which one would you click ? This article is framed towards an audience that doesn't exactly care about technical details.
If you're interested in this area, my company is designing hardware accelerators for these algorithms. Our versions of the algorithms have a few extra twists to them to account for the analog nature of the hardware. I'm probably a bit biased, but I think this is a really interesting place to be working right now
Below is what posted on this month's Who is Hiring:
Isocline - Austin, TX - Software Engineer for High Performance Computing and Modeling
We are looking for two people - one interested in neural networks and one interested in GPS.
We are developing microchips that yield a 10-1000x improvement in performance & energy-efficiency compared to digital ASICs, GPUs, and FPGAs. We are a bootstrapped company and are fully funded through mid 2016. Patents pending.
$70K – $120K Salary
0.5% – 1.2% Equity
Full Job Description: https://angel.co/isocline/jobs/38767-software-engineer
Company website: http://isoclineengineering.com/
Email me directly if you do not have an AngelList profile
https://www.coursera.org/course/neuralnets