spooningtamarin

Realistically, Uber is just an idea, and they got the money mostly for the idea. Even today Uber doesn't have advanced algorithms, Google had that immediately.

Uber is just now trying to solve those hard logistics problems and their prices depend on it.

Google had its own problems too, and someone like Jeff Dean and the team solved most of them.

Ideas take time to materialize, Uber would be nothing without the map data of today, Uber would be nothing without OSM open-source implementation of Contraction Hierarchies.

It's only later when the company had luck raising so much money that they bring in the teams capable of solving the real problems that an idea didn't even know about.

Realistically, Uber founders are good businessmen, not engineers, and not idea executors. This fact alone has made their idea worth less, and even years after Uber started you can outperform Uber.

Uber for Food, with a proper implementation of dynamic vehicle routing problem algorithms can kick Uber's ass.

Unfortunately for those who are not idea executors, the science, the research behind those logistics problems is still immature, and to make "Uber for Food" idea work, you have to have a large knowledge of computer architecture, above average problem solving capabilities, and a pretty darn fast implementation of those algorithms.

That's something Uber can do now, after they employ excellent people, but the problem with large companies is that after some time, they lose direction, they slow down, and that's why you can copy them.

Look at Homejoy, "Uber for Cleaning Services". Their ultimate challenge was exactly the same that Uber has, logistics algorithms. They failed at proper fast implementation, and failed ultimately. Uber has more money, thus more momentum, but they can still fail, because idea is just an idea.

As much as you need to be a very good business man to build a business that Yelp is, you need to surround yourself with good people, and recently, plenty of good people have left Yelp, and realistically, there aren't many out there.

Off the top of my head, checkout work by Eric D. Demaine. His work is on problems that seem very real, and solutions (algorithms) definitely show very lovely interaction of several different areas of research, problem solving skills at its finest.

http://cstheory.stackexchange.com/questions/6660/polynomial-...

Sure. They don't, but when they are just starting school they still don't know much about what is possible, and out there.

> No one is talk about giving them freedom to do what they want.

Little bit of guidance won't hurt. There is nothing intrinsic in child's wishes. No one is blocking their inner genius by guiding them to pay attention to learning maths instead of painting, or dancing, or vice versa.

It is up to the parent to give direction to their child, not pressure them into something that they don't want but show them what is possible and out there.

Giving a child the freedom to do what it wants is probably not going to end up well in majority of cases.

    f(n) = f(n-1) + f(n-2)

I'm not exactly sure why you want your comments to sound like you're a complete lunatic.

This is not a problem with "functional style", it's a problem with library design.

writing

    map (+2) (map (+1) (map (+3) xs)) 

Quantum cryptography has some pretty lovely nature tricks that enable no chance of spying/hacking, measuring the state plays so weirdly with the quantum system.

When you see side-by-side for these upscalers, it's obvious they aren't that good in deterministic upscaling.

Imagine they stored all of the movies they distribute in 4K in your TV. Whenever the movie is displayed they just find it in the database and reproduce the 4K version.

Of course, that is very much infeasible, it requires too large of a storage.

What they do is something like Waifu2x I linked, they train a neural network to learn how to properly upscale movies. They put the "algorithm" on the chip, and its fast enough for real time usage.

They won't tell you it's NNs but it is. Sony distributes a lot of movies, they use their database of movies to train the upscaling models (which is obviously NNs https://github.com/nagadomi/waifu2x ) and then put the chip in the TV.

It's something almost equivalent to storing Pride and Prejudice and Zombies in your TV in 4K, and then reproducing it when they match it with what's playing on TV.

Any algorithms working with images need FPGA implementations to be quick enough, and there are a lot of convnets in use there.

There's a chapter on convnets here: http://www.cambridge.org/us/academic/subjects/computer-scien...

RNNs are nothing special in particular, at least these with small number of layers and nodes.

HMMs or CRFs are easier to handle for sequences, and would probably work well, and there are FPGAs all over the place of these models.

Love this one as a look into the future.

I do not like the research. It is bad, it's rarely reproducible, source code is almost never public thus time/convergence measurements are meaningless.

If your capital isn't allocated fairly results will be pretty bad.

On the other hand you have TSP which murdered all those GA algorithms with a proper Lin-Kernighan heuristic implementation.

I'm not sure if we are talking about same problems. > Saying it's not brilliant

I said they were simple and that's why people like them.

Scientific contributions are mostly in operators, I guess. I can't think what else can someone working with genetic algorithms contribute because adding elitism, or 3-way crossover isn't really the solution to a problem, it's just metaheuristics jibber-jabber.

Same goes for heuristics being faster than exact solutions for harder problems.

I bet that thinking about the problem more would allow you to find an algorithm that works better than genetic algorithms all the time.

Just look at travelling salesman problem, or vehicle routing problem. Smart heuristics beat these hybrid-genetic-memetic-algorithms all the time.

Yeah, if high-performance is something faster than Haskell I'm already there with C++. If I understand copy/move semantics I won't get bitten by anyone, same thing with laziness in Haskell, although sometimes I stumble a bit.

Of course we're all talking about C++14. I guess everyone else is talking about GHC 8.+, at least in this thread.

> You can't really compare development time though can you, even with C++x11? Even if so, you should probably specify the version of C++ you are referring to since they can mean very different experiences.

Write a highly performant discrete optimization in Haskell using monad transformers that have overhead, laziness that has overhead, and almost everything has several levels of indirection. Yes, I can write a prototype that is sufficiently fast, I can then profile and write a version that Haskell might never reach. And yes, it'll probably take me less time to do than to do the same with Haskell, it might be my lack of expertise, who knows, I've tried.

Although, most of the time laziness is only useful at a single level, and that kind of laziness can be easily done in languages like C++ with a much lower overhead.

It's a bit sad that GHC has so many advanced code optimization methods but performance is still lacking.

Yes, you can write high-performance Haskell but it requires expertise or huge amount of profiling and time.

On the other hand, you can easily write faster (speed-wise) prototypes in C++ and get superior performance almost all the time.

List of ways to write performant Haskell: https://news.ycombinator.com/item?id=10209798