As a long-time LaTeX user, I'm a little disappointed to read these results, but I fully believe them.
I've spent countless hours fighting against the layout algorithm to get images to go in the right place, among other frustrating issues.
Efficiency aside, using LaTeX is expected in fields like physics and math, and if you write your paper in Word, readers will be biased against it (consciously or subconsciously).
On the ArXiv, the vast majority of papers are typeset using LaTeX, and of the non-LaTeX papers, a large fraction of them are low-quality or written by cranks, hence the negative association.
> if you write your paper in Word, readers will be biased against it (consciously or subconsciously)
I've found that happens mainly with papers that are not only in Word but formatted a bit weirdly. Those can be avoided by people who use Word regularly, though, and in that case you usually have to look pretty closely to tell if something was done in LateX or Word, if they both use the same template (e.g. the Word vs. LaTeX versions of the ACM paper template).
Telltale "Wordisms" that I run across fairly often, and probably do have a negative reaction to: 1) large spaces due to justification in two-column formats without hyphenation (solution: turn on auto-hyphenation); 2) a paragraph being in a totally different font or font size from those around it (solution: paste without formatting); and 3) PDF title set to something like Paper.docx (solution: set a title when exporting to PDF).
>I've spent countless hours fighting against the layout >algorithm to get images to go in the right place
This probably means you were using it wrong :) LaTeX almost always knows better than you, and you should trust it. You are much better off "guiding" the image placement rules than trying to set them manually!
It's worth noting that embryo selection might be coming really soon http://nautil.us/issue/18/genius/super_intelligent-humans-ar...
and that will have a much bigger influence on the evolution of human intelligence than any effects related to natural selection.
What worries me much more than dysgenic effects is the fact that people in western countries tend to have a strong aversion to any discussion of genetics and intelligence, which means that these countries may end up being late adopters of embryo selection or genetic engineering technology.
- Re: "buzzwords...references": I don't see any buzzwords, in fact the word "deep" doesn't even appear in the text. Regarding references, A typical conference paper references cites a bunch of related papers written by people who might be reviewing it. This paper, on the other hand, cites some seminal work from other fields, which is more interesting and enriching for most readers.
- Re: point of the paper. How to design a learning computer that can access a long-term memory storage of large capacity, which can be optimized by gradient descent. (I.e., everything is differentiable.)
- Re: number of training examples is huge. Training neural networks often takes a huge number of iterations, and the problems considered in the paper are numerically challenging so the iteration count is not surprising. Also, just like the regular Turing machine, the "neural Turing machine" isn't the most efficient architecture, but it's conceptually the simplest one that has the desired properties.
Another robot researcher agrees.
Classifying images is totally different from enabling a robot to perform a difficult task.
Object recognition is a supervised learning problem, mapping image -> vector of probabilities.
Robotic manipulation is a control problem, mapping a long sequence of images -> a long sequence of actions.
"typical incarnations of which consist of over 100 layers with a maximum depth of over 20 parameter layers)"
Anyone know exactly what that means?
I'm guessing that that there are 100 layers total, 20 of which have tunable parameters, and the other 80 of which don't--e.g., max pooling and normalization.
About a year ago, I got to play with a UBR-1 prototype. It seemed to be very well-designed. It was smaller and lighter and the PR2 and had some design improvements. For example, the base was more maneuverable, and the torso and grippers were much faster. The main appeal of the UBR-1 was its low cost -- 50k for the UBR1 vs 400k for the PR2, or thereabouts.
Some of the most technically innovative robotics startups were acquired by Google last summer. Unbounded seemed like the only company left in the realm of mobile manipulator robots, who could continue Willow Garage's legacy of providing technology that runs on open-source software (ROS), is open to tinkering, and benefits robotics researchers. (And plenty more research is required before robots are intelligent enough for the vast majority of menial tasks.) I hope that Google will give back to the research community, but I won't get my hopes up, since they've been extremely secretive so far.
So I'm quite sad to hear the recent news that Unbounded is shutting down, and I'm hoping that the excellent work of Melanie et al. won't be buried due to the legal issues they're facing. (I have no inside info about what is going on.)
Well there are other companies that are making manipulators. Kinova for example looks promising with a polished robotic arm. They don't have the mobile base tough, which is sad.
An integrated mobile manipulator is a very different beast from a standalone manipulator bolted to a generic mobile base.
For example, UBR-1's lift torso means that it can pick an object up off the ground and place it on a tabletop. The placement of sensors is also key, with having a depth cam ideally placed for both looking around and supervising a grasp operation.
Efficiency aside, using LaTeX is expected in fields like physics and math, and if you write your paper in Word, readers will be biased against it (consciously or subconsciously). On the ArXiv, the vast majority of papers are typeset using LaTeX, and of the non-LaTeX papers, a large fraction of them are low-quality or written by cranks, hence the negative association.