New Startup Sets Out to Bring Google-Style AI to the Masses
wired.comHaving examples just executable from the website is invaluable, especially to excite more people. For fun:
Relatedness of Sentences (beta)
1 - not related at all
5 - an almost perfect paraphrase
Two men are taking a break from a trip on a snowy road
Two men are taking a break from a trip on a road covered by snow 4.05
Two men are taking a break from a trip on a road covered by rocks 4.13
Two men are taking a break from a trip on a road covered by mushrooms 4.23
Two men are taking a break from a trip on a road covered by hobbits 4.27
Plenty of work ahead :)
Several children are playing with the leaves
Several leaves are playing with the children : 4.26
Indeed!
Yea, I also wasn't impressed:
"a few guys are drinking water at the mall" -> "stuff" : 4.22
lol
> “deep learning,” teaching machines to recognize images and understand natural language using software that operates a bit like the networks of neurons in the human brain.
'understand natural language'?
Very far from it...
> I saw the movie, where the main actor's wife was so angry - but I was having a great day
Result 99% negative...
> he was hit hard
> he played a hit single
Relatedness 4.5, from 1-5
A solid group of talent, welcome to the club. I am interested to see where the "deep learning" start-ups end up in ten years time with such a wide-array of problem sets and industries.
Am I the only person who is worried that AI start-up companies are going to use AI to insert "deep learning" into every pitch deck possible?
Well, you can somewhat discriminate between them. Just look at people who work at that startup. Have they been doing GPU / Deep Learning staff at around 2010, when it had started showing promise? (this group of people is probably limited to a bunch of grad students and some small communities around a few open source projects). If yes, then you don't need to worry. After all, they've been smart enough to pick up deep learning half a decade earlier, before it had became a buzzword. And now they probably have experience in the area. A solid bet.
On the other hand, if these folks are just some bozos, who picked up a buzz-word and now are trying to hack some stuff together. Well. It's your call.
(I'm not in any way affiliated with that startup. and I haven't checked on the background of that PhD guy. Although I do have some small vested interest in the field of "deep learning" in general. And I don't want it to become yet another dead buzz word. And WTF - why Google is being used in the headline?)
Deep learning has become a buzzword that every CEO on the planet is going to put in his/her presentations.
e.g
Deep learning will allow us to identify new opportunities and capture more value from existing markets..blah blah blah.
is this the sexy re-brand that Machine Learning has been waiting for?
(honest question.. are they the same thing?)
Machine Learning itself is kind of rebranding to avoid itself from being called yet another AI trick. It is AI as much as Deep Learning is. Main difference between traditional machine learning and deep learning is that, deep learning (neural networks) is inspired by biology.
They are closely related but entirely different approaches. Most machine learning is relatively simple statistical models. Deep learning means ridiculously large models. Sometimes they have millions of parameters and require rooms full of GPUs running for weeks to train. But the capacity means they can learn far more complicated functions (like machine vision or language.)
Machine learning isn't an approach, its an entire discipline. Deep learning is just a specific category of implementation of a subset (neural nets) of machine learning.
They aren't entirely different approaches, considering deep learning is a form of machine learning...
Real deep learning[0] is a very particular type of machine learning that has recently been shown to be quite useful for certain specific tasks. But as often happens with technical terms, it's been regularly abused once it started getting press.
Which is not bad.
At some point people will decide it would be good to deliver something with that claim.
Nah, why would most people waste their time worrying about that? If start-ups do or do not, it doesn't matter to most of us. Are you a VC?
'Deep learning' is the new 'big data'
Well, they are just trying to disrupt the big data space with enterprise level deep learning allowing better branding position and lessening inertia to allow quicker pivots. It makes sense to do this because we are at a inflection point where an autonomous thought leader can appropriate the mind share of the by market by leaning in to this space and bringing to bear value added innovation. Also, hacking.
(passes out due to buzzword-itis)
My firm offers the same service, but in a secure cloud.
My firm offers an entire framework built around this service, except it is an enterprise version for 12000% more where the server is placed back into your company's data center.
'Big learning'
I'm not worried, but the term "deep learning" is indeed lame. What does it even mean?
It is "an approach to AI based on enabling computers to learn from experience and understand the world in terms of a hierarchy of concepts, with each concept defined in terms of of its relation to simpler concepts." [1]
Having spoken to Sven a few times, I think they are targeting industries/applications more specific than their website appears :)
If anything, I would consider this to be a competitor to Context Relevant and Alchemy API instead of Clarifai.
Wait til the machines rise against us