My guess is this is the background research for a future product something like google glass for blind people. Leveraging 1980s text adventures, fed thru a speech synth.
"You are facing north in the center of the sidewalk, toward the intersection of main and 1st streets. Standing 40 feet in front of you is a brown dog and 72 feet in front of you there is a woman standing on the other side of the road. Twenty seven feet ahead is the front door of mcdonalds. Would you like to hear the latest google locations reviews of that restaurant? Your apartment building entrance is 157 feet ahead. 34 feet ahead and to your left a graffiti artist has tagged a brick wall with the QR code pointing to the goatse website and there is a billboard with an advertisement for the latest star trek movie to your right and 50 feet upward. Also it is dark and you are likely to be eaten by a grue."
This is the paper that did unsupervised training of a deep net on frames from YouTube videos, and found it had autonomously developed detectors for, among other things, human faces and cats. Jeff Dean is a coauthor.
Such research could also be of immense value to sighted people - automatic, continous object recognition will be very useful for augmented reality applications in a Glass-like device.
- Google+ queues images for recognition. Results improved steadily over 72 hours.
- Google+ does not use OCR of text in the images. That surprised me. But perhaps it's a privacy issue.
- Google+ does use information gleaned from elsewhere on the web. Words that were associated with the same images on Flickr would turn up those very pictures on Google+.
- Oddly, Google+ does not use information associated with those images on Twitter.
- Google probably uses EXIF data married to a database of location names.
- The much-vaunted feature recognition is impressive, better than any other system, but for me did not achieve creepy levels of intuition.
And it also doesn't seem to stem words. [flower] and [flowers] give different results (actually flowers gives no results). But I am impressed by the number of classes they have: who labels pineapples in an image corpus?
The unlabeled object recognition test is a standard test of machine learning algorithms.
Historically, error rates of around 20-25% won competitions and set records. A year or two ago, though some researchers and professors from the University of Toronto absolutely smashed those records, getting around a 16% error rate. They went and made a startup out of their tech, and got acquired by Google a few months ago.
I think that this is going to be the first of a long line of Google products integrating this sort of deep neural network technology. I wouldn't be shocked if Google in 10 years was known for something besides search, at this rate.
Wow, that's awesome. Imagine your photo collection of 1000s of photos, and you remember "the one with that cat" but how do you dig through the photos to actually find it? This can go a long way to a much more meaningful photo library management experience.
Am I the only one scared by the thought of uploading my whole photo collection to Google's servers? What about creating an offline database of object fingerprints that can classify my pictures without privacy violations?
Anyone's free to build that product. Of course, not anyone is capable of this - surely Google leverages immense computing power, complex software and its knowledge graph to do this analysis? You'd need to replicate all that in competition.
But seriously, how paranoid can people be? If anyone really wants to get your data, do you really think it's safe on your server or on your local machine?
By the same token a really determined burglar can get into my house - that doesn't mean I should leave the doors unlocked.
No-one is suggesting that Google are going to hack into your machine to get your data, nor that what they want to do is out and out unpleasant, but what it is in in their interest either instead of or as well as yours.
Until we work out that instead of / as well as, I think a healthy questioning of what might be happening is reasonable.
Yeah, it would be great if you could download the photos with the recognized keyword tags embedded into the EXIF information. I really hope they do this.
Google has been building her knowledge graph for a couple of years. The goal is for computer to truly understand real world concepts rather than keywords and text. I didn't fully understand the application of it rather than some fancy cards on the search results page until yesterday when I asked Google "where did Golden retriever originate?", and Google answered "England". Google might not really understand the concept originated or golden retriever, but Google understands that "where" is asking for a place and she found a lot of mention of "England" in all the page results of "golden retrieve origin", she also understand that England is a place. So Google guessed the answer.
The Google computer has been reading about these concepts for years, now we know it can see them in pictures (and maybe even in live videos). That excites me to a degree that it becomes a little bit scary. When will that computer learn the concept of "self"?
Update: actually Google seems to understand the concept of "golden retriever", I search my photos with the word and yes, at least Google knows how golden retrievers look like.
We actually have an explicit concept of Golden Retrievers and their origins as an Animal Breed within the subset of the knowledge graph that we expose with a permissive license: http://www.freebase.com/m/01t032
The data is available for querying as well as licensed such that you can take it and build your own commercial database with it (requiring only attribution).
Self preservation is just a selection function. I personally think that all it'll take is a mutating algorithm that can pay for its own hosting by somehow acquiring money and a self preserving algorithm like human intellect could eventually emerge
What if it emerges? For instance, the computer can be programmed with the notion of "doing things with or for" something. What do you do with or for a self? Preserve it.
Our instincts emerged from billions of years of natural selection, and they work at a much deeper level than our logical thinking. The idea that intelligent computers would start from logical thinking and then develop human emotions and instincts doesn't have any basis besides our natural - I would say instinctual :) - tendency to anthropomorphism.
I don't think so. The neural network in Google+ was trained on labeled images and now finds similar objects in unlabeled images.
The technology discussed in that article is about deducing the existence of a common feature, in this instance a cat, from a large collection of unlabelled images.
It may be the same tech (roughly). Use the same approach for all but the last layer, then use traditional backprop to learn the last layer and fine tune the connections in the lower layers.
Mostly unlabelled then, which means you can learn to generalise over a huge number of images but learn labels on a smaller set.
Yep. I don't know if that's what is actually being used here, but that is pretty much how they did it with the same system:
"We applied the feature learning method to the
task of recognizing objects in the ImageNet
dataset (Deng et al., 2009). After unsupervised
training on YouTube and ImageNet images, we added
one-versus-all logistic classifiers on top of the highest
layer. We first trained the logistic classifiers and
then fine-tuned the network. Regularization was not
employed in the logistic classifiers. The entire training
was carried out on 2,000 machines for one week."[1]
Basically you learn features in unlabeled data, then identify the features your trained net is recognizing with labeled data. When you run over g+ images, you then only tag with features you're sure of past some threshold of certainty.
I just tried it on my photo collection and it's incredible. It even works for famous places, e.g. I searched for "western wall" and "dome of the rock" and it found them. I can't imagine how that works
There are tons of photos of these places online, many of them tagged ("breaking news from the dome of the rock", or "here's me and Sam at the western wall"). Collect enough of these and you can attach knowledge to images. Then, you just have to know two images look similar, and you have your classification.
Neither of the above is easy - nay, it's very hard. But once you have those two building blocks, this technology is viable. And it's very exciting!
I took a half dozen or so pictures of the sagrada familia, all have gps data in exif. Only one of my pictures contained the famous spirals, the rest were closeups of exterior detail. Only the picture of the spirals showed up in the query.
My photos don't have GPS positioning. I did manually position them roughly though, but nowhere near accurately enough to know which photo is a specific landmark and which isn't
I think the greater achievement came with Google image search, someone had to tag all those photos.
They wrote an algorithm that takes that data and recognises new images with it. As long as there is a way for us to tag inaccurate matches then it should be able to continue to learn. I imagine any flagged matches are being reviewed carefully.
I always thought that was done using keywords on the page the image was taken from (and image captions, alt text, titles and filenames). This is reinforced by what you see when you go many pages ahead in images and see things that don't seem related to what you searched, but the keyword is there somewhere on the page.
This is seriously fun! You can actually search for "blue car" and it works. Searching for "picture" results in an error however. Same for "image". "photo" seems to return more or less everything.
Object recognition also works on videos, judging from the fact that a recording of my cat came up in search results for “dog”. (Could be that it only looks at the first frame, though.)
"You are facing north in the center of the sidewalk, toward the intersection of main and 1st streets. Standing 40 feet in front of you is a brown dog and 72 feet in front of you there is a woman standing on the other side of the road. Twenty seven feet ahead is the front door of mcdonalds. Would you like to hear the latest google locations reviews of that restaurant? Your apartment building entrance is 157 feet ahead. 34 feet ahead and to your left a graffiti artist has tagged a brick wall with the QR code pointing to the goatse website and there is a billboard with an advertisement for the latest star trek movie to your right and 50 feet upward. Also it is dark and you are likely to be eaten by a grue."