I used to think I was the same, but I realised (embarrassingly slowly), that every time I went camping/hiking away from artificial light I would just go to a natural sleep cycle. It was artificial stimulation all along, I wasn't really a night owl.
As a counterpoint, Neptune and Pluto were "dark matter" prior to confirmation with observations. They were theorized to exist due to their perturbations to observed motion of known planets.
There are two frequencies (sometimes three) transmitted by the satellites. Using two frequencies allows certain atmospheric delays to be accurately estimated. Trouble is only one code is available for civilian applications. The trick with two receivers is to solve for the phase of secret signal (rather than decode it) by solving an integer least squares problem. This allows accuracy of the order of +/- 5cm
Brings back memories of my friend borrowing the lab computers over the weekend to do some long running Monte Carlo simulation runs (so trivially parallel). Except he had a bug that implicitly gave all the runs the same RNG seed. Took a bit of head scratching to work out what the hell had happened.
I don't think bending modes are the usual culprit, it is more often the frequency mismatch between axial and transverse modes that cause inharmonic tones. I'd need to get a real piano and make some spectra and do some maths to show it though.
Also your network infrastructure is much cheaper than a lot of places around the world. This is a large part of the price of electricity where I come from (which is far more expensive).
That really depends, a lot of people generate too much solar during the day and typically feed it back into the grid for peanuts (especially in regions with high solar uptake, the wholesale cost of electricity gets towards zero in the middle of the day). Maybe more people doing this instead of feeding back into the grid in these places makes some sense.
It really depends on an individuals displeasure at work though right? To take a ludicrous example my utility function for eating a sandwich is has a sharp discontinuity as a function of how much Polonium is in the sandwich. So too with some people for money and work, especially when it is far away from your current state. I don't think this is an experiment that a real economist (TM) has ever performed on an individual to discover what the true utility function is (happy to be wrong though).
I always say that AI is a forever moving goal post. It is simply a task a human can do that you wouldn't expect a machine to be able to do. So as soon as a machine can do it, people no longer consider it intelligent (i.e. it is just A*, it is just a chess engine, it is just a network picking up on patches of texture, ..., it isn't really "intelligent").
This is really an experience thing. You are experienced enough to chunk bits of the work together so that there is no context switch. Beginners can't do this, that's why experienced cooks are so much faster. They intuitively know what will "fit" together in time. This generalises to other tasks...
If you go back in the past, contemporaries of newton were critical of his gravitational theory having an instantaneous action at a distance. Familiarity has just bled us of its strangeness over time.
I think AI always means "does something a human can do that you wouldn't expect a machine to be able to do". So it is forever moving goal post. It just moves ahead in step with people's expectations of what a machine can do. Also magic is always disappointing when you know how it works. We just have the psychological safety of not knowing how we work, whereas in ML we always know what it is doing, and it is always kind of disappointing.
"... the Inertial Navigation System they use while submerged has enough sensitivity to be export controlled."
To further your point, if the inertial navigation system has low limits for max acceleration and angular velocity (i.e. can't be put in a missile) then it can still be accurate enough for this task without ITAR restrictions.
In my experience people will come at this from a variety viewpoints. Typically they (1) don't trust the model to learn something useful, so they want some confidence that the model isn't going to do exotic things with new inputs (i.e. they want some faith in the generalization ability of the model to unseen inputs) or (2) they want the model to help them understand the problem. Your language example is perfect. How nice would it be for linguistics if a complex model could tell you about simple structures that you didn't previously know about. It is nearly an article of faith with some people that these simple structures must be there generating the statistics we see, like they haven't considered the possibility that there might not be a simple structure underneath.
Probably due to the really idiosyncratic way I think about things and from the post it is not at all clear what I mean, sorry for the imprecision! I think of cross correlation of 1D signals with a kernel as sliding one of the functions over the other [0]. In my head I imagine drawing the signal and kernel next to each other one a big piece of paper and literally folding it over and sliding it back over the signal to intuit the convolution (with an imaginary see through piece of paper). I really don't think that was obvious from what I posted, so definitely not dyscalculia!
I might be wrong (so happy to be corrected by a native speaker) but I liked when I heard that the German word for convolution is "faltung" which translates to "folding". This is a much nicer word to metaphorically understand the operation. Maybe it still confuses the hell out of German undergrads though?
"...and hear your stories about how you’ve done this before."
Basically Mr Kurtz saying "The horror! The horror!" gives you the right impression.
For small datasets and short training times it isn't so bad, a deterministic training script on git and a dataset with gitlfs. So can easily reproduce the model when needed.
Having to do this on very large datasets (that are possibly expensive to keep a copy of for just experiments and not prod) or really slow training times is basically disgusting. I'll be interested to see what you have done.
The scaling power coefficient matters, they are so far off our mass that the coefficient might just be wrong. Neanderthals are about the same size so an error in the coefficient won't matter so much.
It is hard to imagine a material strong enough to maintain residual stresses this high! Maybe in diamond? And what happens when it fractures and all comes apart? (unless it is contained in a very small volume so has bugger all potential energy)
Plenty of people watch the tour de france, which is basically low powered slow racing. The strategy involved in exploiting aerodynamic gains in the pack is kind of fascinating. Trying to sell it as the same thing is kind of dumb though.
You just need "distance" in the k means to be more general than just using the euclidean distance between coordinates, you need a norm appropriate for the space you are working in. Haversine distance should be fine for this application.
I've always had this idea that you could do a dumb version of annealing by just progressively increasing the size of samples used in the minibatches. SGD gives you noise on the gradient which should get smaller as you increase the batch size. If you just adjusted the batch size with some sort of MH accept/reject logic you should get a better optimzer for mode seeking than just vanilla SGD. Like all my ideas though it has probably been thought before or is just wrong :) too damn lazy to try it as well.