GrantS

There appears to still be active research about mechanisms that explain the historical data but here's a clearly written article [1] discussing a few factors such as: variation in solar radiation due to variation in Earth's orbit, changing ocean currents which may lead to huge upwellings of carbon dioxide from previously sunken dead algae.

[1] https://blogs.princeton.edu/research/2013/04/30/how-the-ice-...

Interesting analysis, but after reading the FAQ at the bottom, it rests upon quite a few important hidden assumptions. For example:

>However if there is any hope of finding life on other planets, there must be a huge number of planets with life in the Universe. Therefore, for the case we're interested in...

So during the entire analysis, we were limiting ourselves only to those universes in which we do make alien contact, regardless of how likely that event is.

The classifiers were probably trained independently and are running independently here as well, not in a hierarchical fashion.

Another way of saying this is that while the tiger training data consisted completely of tigers (for positive training examples), the animal training data might not have had any tigers at all (however unlikely) -- i.e. it could be recognizing the tiger image as containing similar enough features to the images of dogs and cats and lions that it did see to trigger the animal classifier, but with only 84% confidence.

Good points. Just to clarify in case folks take your experience in 2005 as the rule, while certainly not as widespread as today, ML was definitely being offered at various universities in the late 90s and early 2000s: for example, the textbook "Machine Learning" by Tom Mitchell [1] was published in 1997 and was used in the undergrad ML class at Georgia Tech when I took it in 2001-2002.

[1] http://www.cs.cmu.edu/afs/cs.cmu.edu/user/mitchell/ftp/mlboo...

This is some very cool research, though many may be surprised to learn that it is an IJCV journal article from 2007, based on a conference paper from NIPS 2005.

Source: http://www.cs.cornell.edu/%7Easaxena/learningdepth/

Regarding the synthesized singing that begins 10 minutes into the video [1], I had long ago read that the reason HAL sings "Daisy, Daisy..." in the film 2001: A Space Odyssey is that Arthur C. Clarke had seen a demonstration of a computer singing this song in the mid 1960s, but this is actually the first time I've seen or heard (what I presume to be) that original demonstration!

[1] https://en.wikipedia.org/wiki/Daisy_Bell

You would probably be interested in the computer vision work of professor Nathan Jacobs over the last decade: http://cs.uky.edu/~jacobs/

He has done some amazing work extracting geo-location and scene structure based on daylight patterns, shadows cast by clouds, rainbows, etc in images and image sequences.

Google's Cardboard team is working on Jump [1], which is a 16-camera rig with a computer vision post-processing step to create stereoscopic 360 video.

Lytro is working on Immerge [2], which is a light field camera for "cinematic VR". So in addition to stereoscopic 360, you should be able to move your head around within the volume of the sphere as well.

[1] https://www.google.com/get/cardboard/jump/

[2] https://www.lytro.com/immerge

The interesting data here aren't the news articles themselves, but the news-browsing history of 20 million people over a 4 month period.

To answer your first question, though, according to the official description of the dataset [1], "On the item side, we are releasing the title, summary, and key-phrases of the pertinent news article."

[1] http://webscope.sandbox.yahoo.com/catalog.php?datatype=r&did...

I'm assuming "mAP" in this context is "mean Average Precision": http://fastml.com/what-you-wanted-to-know-about-mean-average...

The other possibility would be "maximum a posteriori" which doesn't fit their usage here.

It's true, but it's such a shame. I still remember the day metacritic switched away from their original design [1] where every movie currently in theaters was in that single column on the left hand side. You visited one page, you looked at it for 15 seconds and you were done. Apparently this was not the best way to make money, but it was so useful.

[1]https://web.archive.org/web/20090117041216/http://metacritic...

The most interesting bit to me was learning why the mission involves orbiting Jupiter rather than orbiting Europa directly. Though the text doesn't make it clear, the video explains (starting at 0:59) that directly orbiting Europa would mean too much time spent in the most intense parts of Jupiter's radiation belts. Making 45 Europa fly-bys apparently avoids radiation-related issues and allows for making observations over an extended period of time to detect changes.

EDIT: The Wikipedia article on Jupiter's Magnetosphere [1] has great details on the impact of Jupiter's radiation on previous missions. For example, Pioneer 11 lost most of its images of Io, and Galileo had total data loss on three of its orbits.

[1] https://en.wikipedia.org/wiki/Magnetosphere_of_Jupiter#Explo...

The end user never sees the raw video in this case so that's not a requirement -- they're generating lots of synthetic stereoscopic views for the user by post-processing the video and accounting for depth.

I'll be interested to see what data they're sending to the video player to provide both stereoscopy and 360-panoramic video at the same time. The methods I can think of offhand require real-time pixel-shuffling to assemble the proper view during playback but maybe they figured out something else.

Right, after playing through a whole song on the web, my first reaction when I saw that there was an app version available was: Oh, that's brilliant, the web app (which I thought was the end of the story) is actually an interactive demo for a paid app. So I was surprised to see that the app itself was free -- but perhaps this is the first time the web version has gotten any traction, so the developer was unprepared for customers with this set of expectations.

Interesting. There must be something unique about the market for apps in Norway as I managed to hit a similar milestone [1] (outselling Angry Birds in Norway for one day) in 2011.

[1] http://trimensional.com/blog/2011/05/23/trimensional-is-1-in...

The direct contact for rides and purchases is there to prevent accidental activation -- you have thousands of people walking around with these bands in close proximity. In contrast, when pulling up to the gate at one of hotels, you just wave your wrist a foot away from the sensor out of your car window and it turns green and opens the gate -- no awkward stretching required.

Thank you for the link, that was a wonderful essay. And you've just reminded me that I've had an old used copy of Clarke's "Profiles of the Future" (the apparent source of this article) sitting on a shelf in the next room unread. Time to change that.

The distance threshold was determined by hand on a per scene basis, and then we find connected components in the graph of neighboring 3D points, so it's a bit of a hack, but there are two important things to note. First, we can get away with an overly permissive threshold because we also require that any two grouped points are also observed at the same time in at least N images -- meaning SIFT features were detected for both 3D points in the same image (so they therefore exist at the same point in time). This filters out lots of spurious groupings. Second, this approach was inspired by super-pixels, which is an OVER-segmentation of an image into groups of pixels that are somewhat coherent -- each super-pixel probably lives on the same semantic object in the world, but they are by no means complete. Still, it's massively better than reasoning about individual pixels (or individual points).

So we err on the side of dividing single buildings up into multiple semantic objects. If our data included detailed reconstructions of the streets between each building, then the whole thing might be connected and we'd need more criteria to separate them out -- we do automatically estimate a ground plane so that's one way: just ignore everything near the ground for grouping purposes.

There's slightly more detail in the paper: http://www.cc.gatech.edu/~phlosoft/files/schindler10cvpr.pdf

So for anyone interested in this, I did my PhD thesis in computer vision on automatically reconstructing 3D models of cities as they change over time purely from historical and modern photos -- you can see my results for lower Manhattan (1928-2010) in these slides from CVPR 2010: http://www.cc.gatech.edu/~phlosoft/files/schindler10cvpr_sli...

We were also able to estimate the date of historical photos fairly accurately by first incorporating them into a 3D reconstruction and then reasoning about the visibility of structures in the scene.

Some videos of the time-varying 3D point cloud here: http://www.cc.gatech.edu/~phlosoft/

That page also has an interactive demo of a unified 3D model of Atlanta from 1864 to the 2000s based purely on image-based reconstruction. Here's a photo of Atlanta from the civil war (1864) with a modern skyline rendered into it by projecting the reconstructed building geometry and textures from modern photos into the recovered camera position of the civil war era photo: http://4d-cities.cc.gatech.edu/atlanta/img/atlanta1864_2008....

Official project website with video and earlier publications: http://lis.epfl.ch/DALER

Official press release from this week with more details on the recent paper: http://actu.epfl.ch/news/a-flying-robot-that-can-walk-4/

It's amusing to note that the primary author of the paper (Ludovic Daler) created a meaningful acronym for the project that matches his last name (DALER: Deployable Air-Land Exploration Robot).

It's interesting how they go out of their way to describe this as NOT augmented reality when... that's exactly what it is. The only time the term appears on the product page is here:

"Microsoft HoloLens goes beyond augmented reality and virtual reality by enabling you to interact with three-dimensional holograms blended with your real world. "

I understand the marketing reasons for this, but contrast this with the fact that Oculus embraces the term "virtual reality" despite the baggage that comes with it and the fact that they can't trademark it. I guess AR never caught the public imagination like VR did.

It's interesting to note how much has happened in the short 891 days since that previous discussion occurred in mid 2012. At that point, the space shuttle fleet was retired but SpaceX hadn't yet flown a commercial resupply mission to the ISS. Curiosity hadn't yet landed on Mars. Rosetta/Philae arriving at their comet, etc. I'm sure someone can come up with a more complete timeline, but I've been following space exploration as a layman for a number of years and these last couple of years have been some of the most exciting in quite awhile.

Ha, I was thinking along exactly the same lines when I read that paragraph. Last week, I found myself reading a 1784 magazine article [1] about "Aerostatical Experiments" (the first hot air balloons) which referred to "inflammable air" which is what they called hydrogen in those days. Google n-gram viewer gives a beautiful illustration of when the name changed [2] -- this is an obvious switchover but I imagine many words change meaning and usage more slowly and in more subtle ways, so your proposal was quite exciting to think about. Let's hope someone takes up that line of research, either from the humanities side or the machine learning side.

[1] https://books.google.com/books?id=lvsRAAAAYAAJ&lpg=PA29&ots=...

[2] https://books.google.com/ngrams/graph?content=inflammable+ai...

He makes an excellent point about the possibility of comparing word vectors trained on different corpora to make quantitative statements about differences in culture, either over time or between sub-cultures:

"I’d like to emphasize that which words are feminine or masculine, young or adult, isn’t intrinsic. It’s a reflection of our culture, through our use of language in a cultural artifact. What this might say about our culture is beyond the scope of this essay. My hope is that this trick, and machine learning more broadly, might be a useful tool in sociology, and especially subjects like gender, race, and disability studies."

Correct -- it is a completely untethered, mobile VR headset and for now has no positional tracking at all. The entire motivation for GearVR existing and Carmack putting all his time into it is to accelerate the arrival of a standalone VR headset based on Android. Carmack seems to have transformed into a computer vision person the last few months in order to implement inside-out markerless tracking for future iterations. So view this as a parallel development track to the Oculus Rift, where the rift is focused on the best experience possible and the GearVR is focused on being an all-in-one device.

I had a similar reaction. In addition to the article's assumption that the consumers and producers of audio are distinct groups with no overlap, there is the assumption that humans will be the only ones listening (rather than computers, e.g. Shazam).

Those two factors are probably not enough to change the way ALL music is distributed, but they deserve to be acknowledged and dismissed with proper evidence as well.

I ran this on a Mac, should work on Linux as well. Since it may not be obvious, the "blender" command line executable is located within the regular Mac OS X Application bundle for blender that you would download from the web:

/Applications/Blender/blender.app/Contents/MacOS/blender

To follow up with a working example of python via headless blender on the command line, the following 4 lines interactively get me an image of a sphere saved to /tmp/.png

./blender -b --python-console

>>> import bpy

>>> bpy.ops.mesh.primitive_uv_sphere_add(size=3.0,location=(0,0,0))

>>> bpy.ops.render.render(write_still=True)

I have experience running blender in a completely scripted manner, from geometry creation, to camera motion, to rendering. These links may help:

(1) Blender 2.6, Python Manual with full API and Tutorials

http://wiki.blender.org/index.php/Doc:2.6/Manual/Extensions/...

(2) Run python scripts through blender from command line:

blender -b -P script.py

blender -b --python-console (for interactive console)

http://wiki.blender.org/index.php/Doc:2.6/Manual/Render/Comm...

Microsoft Research published an early version of what you describe in 2004, "Image-Based Interactive Exploration of Real-World Environments":

Paper: http://research.microsoft.com/en-us/um/redmond/groups/ivm/iv...

Web Page, Video, Slides: http://research.microsoft.com/en-us/um/redmond/groups/ivm/iv...

Quoting: "Whereas many previous systems have used still pho- tography and 3D scene modeling, we avoid explicit 3D reconstruction because it tends to be brittle."