jarvic

There is a difference between "never been shown to" and "shown to never".

Interestingly, if you remove "rund" it changes from "only 50 yards" to "just 50 meters".

If memory serves Opera was the first browser that had tabs but they were also in Firefox by the time Chrome released.

As someone who was in one of the best low-sec/0.0 pvp corps for several years, I completely agree with you. Theres no other game that quite matches the feeling of dancing around each other until someone manages to pin an important enough target down long enough that both sides have to commit to the fight. PVP between small to medium sized gangs in EVE is the best PVP I've ever participated in.

You indent line A with tabs. You use tabs to line the start of line B up with the start of line A (indentation) and then use spaces to move it the rest of the way to where you want it (alignment).

Comments that take legitimate concerns with the merits of a particular study and dismiss them as misogyny don't add anything to the discussion either.

I run MATLAB on Ubuntu 14.04 every day and have never had any of the issues you're describing. Are you sure there isn't something else going on there?

It almost killed me when mine broke earlier this year and I found out they don't make them anymore. Best mouse I've ever used, and its corpse still has a place of honor at the top of the pile of old hardware in my closet.

Join a PvP corp, fly cheap ships, do just enough missions/exploration/pirating/trading/mining/whatever to keep yourself in ammo, and have a blast shooting people. EVE only turns into Spreadsheets in Space if you make it that way.

I agree that the title could be read both ways, but as someone who works with CV, I knew immediately what it meant. CV outside of a few specific applications on controlled images is only just starting to work at all.

Yeah, I wasn't trying to be disagreeable, just pointing out that this is a very active area of research still, and there is always room for improvement. I work mostly with segmentation and shape analysis instead of registration, but I am involved with a project on multi-modal image fusion, which is even harder (and more interesting).

The trick with overlaying follow-up scans is called image fusion and is easy and can be done by one-click applications

Just wanted to point out that image fusion is far from easy. The brain is easier than most areas of the body because the skull provides a good basis for performing a registration, but even then the brain has some small room to shift and you can't get a perfect alignment with only rigid transformations, especially when you're dealing with tumors.

For most other parts of the body you have complications like organs shifting around, gas passing through the digestive system, patients being in slightly different orientations, etc. In these cases you most definitely need to use some kind of deformable registration, which is far from a solved problem, especially when you have things like tumors changing shape/size or even appearing from nowhere from one scan to the next.

The issue is when people (seemingly) randomly switch between he/she and his/hers, even within a single context. Then you end up with something like:

"If your child is in the kitchen, make sure she doesn't touch the hot stove. If he does, however, ..."

Even when the switch isn't that abrupt, if someone is switching between he and she every few paragraphs, it can become very distracting. Using singular they solves this problem.

This is also one of its major drawbacks, though, because the features it "discovers" are often not easily interpretable. Sure, you can get good classification performance, which is sometimes good enough.

I work with a lot of scientists, though, who don't just want performance, but insight. This is where neural networks and related techniques still fall short compared to more "traditional" methods.

There really is encouraging progress being made in this area (early prediction of AD) in several different fields. I know of several different groups who are using imaging instead of blood tests to predict conversion of MCI to AD as well. A lot of it is using the data being collected by ADNI:

http://www.adni-info.org

I haven't seen anything specific about how they are doing it, but I can give you a few guesses as to what common approaches would be.

The most straightforward way to model the shapes themselves is to put some points on each letter in corresponding locations on each letter. Every sample of each letter would need the same number of points, and each point would need to be in roughly the same spatial location on the letter (so, for a letter S, the first point in every sample would be at one end, the last at the other, and the ones between at some kind of identifying landmarks).

Given these collections of sampled curves, the simplest thing to do is to just compute the Euclidean mean, treating each of them as a point in a high-dimensional space. You could go farther and do PCA, giving you not only a mean but modes of variation. Using this you could examine the most common ways in which each letter varies in the population, which can be an interesting thing to study.

What I've described is building a statistical shape model from a boundary point distribution model (PDM) of an object. This is typically done in the context where you want to fit your model to new instances of the object, not just for finding a mean, and are known as Active Shape Models. You can check it out on wikipedia (en.wikipedia.org/wiki/Active_shape_model). Here is one (among many possible choices) describing generally how this is done:

http://tinyurl.com/nxjwygn

There are other techniques for representing the shapes or computing statistics that can produce better models, both from theoretical and practical points of view, but this is generally the most common and would be my first choice if I was going to do something like this. Of course, they could be doing something much simpler like a simple averaging of the images (assuming the letters are all in roughly the same place) as well.

I just skimmed the post as I don't have time to fully read it right now, but I'll point out a couple of problems that you can run into with neural nets and associated approaches.

One issue that can be a back breaker depending on your application is that, to produce a generalizable model, nets tend to need much more training data than the alternatives. There are ways to work around this, though.

The bigger problem to me is interpretability. Deep learning often gives feature sets that are very good for whatever task you are working on, they are in some senses artificial and it is difficult to relate changes in features to changes in the input data. I work with a lot of biological and medical data, and this is an issue because for some applications it is important not to just get accurate classification results, but to be able to understand what your features mean in the context of the original problem. I saw some interesting work in a computer vision paper earlier this year on trying to learn how to visualize how changes in input and outputs of a neural net were related, I'll try to dig that up later if anyone is interested.

I'm not sure how coherent that was as I was trying to get this typed out in a hurry.

Hello, sir. Do you have time this morning to talk about our Lord and Savior, Bird Jesus?

The point isn't reinventing the wheel. It is understanding why the wheel was created, why it is better than some of the alternatives, the pros and cons over using one kind of wheel vs another, and when it is time to stop using a wheel and invent tank treads instead.

Disclaimer: I'm a CS PhD student that did pure math in undergrad, so I probably have a different perspective on these things than someone who did CS only.

CS, at a graduate level, is a much more varied field than most people, even CS undergrads, realize. In fact, a lot of the main research groups in my department (graphics, vision, image analysis, robotics, VR, visualization) are things most CS undergrads don't even get exposed to.

In math or physics, a lot of the undergrad courses are kind of basic, entry-level things that give you a feel for some of the different fields of research without going too far in depth. This isn't really the case with CS, because CS departments have to cater to several different audiences who tend to fall into one degree. They have to come up with a curriculum that balances programming/development concepts for people who want to be developers, networking/administration stuff for people who want to be IT (although this is more and more being moved into a separate program), and theory for people who actually want to be computer scientists. Even then, this theory is mostly relegated to computation theory, which is only a small part of what research computer scientists do.

Personally, I'm in medical image analysis, which is a field which is a collision between a ton of different areas, including physics, statistics, operations research[1], numerical analysis, image processing, differential geometry, and medicine. I don't know that I would consider any of those fields to be the "easiest branch of mathematics" (what does that even mean, anyway?)

This post has kind of gotten off the rails a little bit, so I'll end with this: I think, if you just look at a CS undergrad program and try to project that to what actual CS research is like, you end up getting a much narrower and incomplete picture than if you do the same thing to a math or physics department, partly because CS research tends to be much more interdisciplinary than others (I have many more collaborators in many more areas than some friends of mine in math/physics/biology do).

--

[1] http://en.wikipedia.org/wiki/Operations_research

You need a higher bar than 'learned something' if that something is a negative.

Why?

They don't want to teach any "tools" b/c CS professors find teaching people how to be good developers as beneath them. They honest to god think CS is an actually hard science on par with math and physics.

Despite what you (and a lot developers that I've talked to) seem to think, CS is a hard science. However, CS and software development are not the same thing. Most CS programs are taught by professors who have research degrees in Computer Science, not by people who have ever been (or probably ever claimed to be) professional software developers, so there is a fundamental disconnect between what you seem to think you should be learning and what is actually being taught.

The professional software developer is a fairly new occupation in the grand scheme of things. The people who are most qualified to teach it are people who have actually worked as developers in the industry, but that is difficult because 1) most universities are reluctant to hire (and students can be reluctant to take classes from) adjunct/non-PhD faculty, and 2) it is hard for schools to offer enough to lure many of them away from industry anyway.

Why not use VS? I have plenty of complaints about various Microsoft products, but very few of them are on the quality of their C++ IDE.

Everybody, get your popcorn: https://lists.debian.org/debian-ctte/2014/02/msg00344.html

Jobs specifically claimed that every version of OS X had been compiled for both PowerPC and Intel processors, though I can't remember if any proof of this was ever given.

Sure, most anyone can throw some data into an SVM and get a result out, maybe even a good one. The problem comes when someone like this has to answer questions beyond a simple 90% accuracy rate. What does the computed separation direction tell me? Could I improve accuracy by using some a priori information like how often one class occurs in relation to the other? What 10% of the population am I failing on? Is it an important part? Is there some easy way I could do better? Is my data so high dimensional that I'm getting some trivial separation and not anything driven by the data itself?

And what happens when this person gets a new data set and they are suddenly getting garbage out of some standard SVM? Is it just a matter of the data not being well-separated using a linear model but throwing some simple kernel at the SVM will do the trick?

Even something as simple as taking a mean can fall apart when you are dealing with data which doesn't live in a Euclidean space, let alone something like PCA or SVM which also make assumptions of linearity.

The point is, it isn't just about being able to invent new methods. Things like SVM make assumptions about your data and applying them in cases when these assumptions don't hold can give completely worthless information, even if it looks good on the surface. Using something you don't understand, even if it is at a (much) more basic level than someone with a PhD in statistics, is just asking for trouble.