I don't know, if your generative model is something like getting the knn of each data point, then fuzzing each point by the local statistics of each point (from the knn's), I don't think you will miss the important trends and you get a bit of differential privacy.
"If this were true, that would mean all previous researchers were wrong."
It's poor phrasing, but in a review I would interpret this statement as "I made a strong claim, and my discussion wasn't strong enough to address the concerns of a reviewer who has only interacted with my work via this document." not as "you must be wrong because everyone else says something different".
Reviews are a weird social process though.
It is also possible to travel faster than some waves emitted in the ground. Particularly when going from a rigid to a soft foundation (like coming off a bridge).
The air compression going into tunnels can cause drama as well. Everything gets more difficult the faster you go.
If (and it is a big if) electric power had an advantage, surely they would just have a support vehicle with a big generator that they could recharge from.
For sure for a real gimbal, I'm just saying that you can do the same thing mathematically to avoid singularities when dealing with rotations represented as euler angles.
Gimbal lock is a problem for euler angle representation of rotations, not rotation matrices themselves. You can fix it for euler angles the same way they used to for old mechanical gimbal gyroscopes. By having a fourth rotation of the frame that points the singularity out of the way. This is a typical solution in navigation for dealing with the poles of the earth (a wander angle frame).
This plot suffers from my pet peeve for geographic display of data. Small areas will have all the interesting details (e.g. extreme outliers in inner city areas), and you just can't see them from a zoomed out perspective.
Have they tried paying them directly to not perform muggings? Or better still in programs that reduce the desire/necessity to mug. The current system does not appear to work, probably worth trying something else...
Valid points that are hand wavey, non specific and non quantitative. Basically, you can't estimate parameters because of fat tails so what people are doing is inadequate. Seems trite to me
I am going to be obtuse and say that since Bayes theorem is a non controversial rule about conditional probability, if you interpret your NN as a probability distribution over outputs that is updated by data, it is always able to be interpreted as "Bayesian" and can be a helpful way to examine what your implict priors are (i.e. via architecture or regularization terms) to see if they are reasonable for the problem at hand. It is no surprise that weak uninformative priors are sort of useless. Explicit priors shine when you know something about the actual problem (say some moments or some invariances in the problem set up).
Computing weighted sums and thresholds gives you a universal function approximator, so that's nice. I'm more interested in how the buggers set those weights and thresholds. It is hard to imagine them having access to any gradient information, so how do they work?
That is beautiful. I think Judea Pearl has been a little too successful in getting people to question causality without being successful at getting people to think about causal reasoning.
Planned burns are a thing here in Australia. It has turned into a bit of a populist talking point amongst certain people. The truth is that the time window to safely do burns in winter has sort of disappeared. Plus these fires have been intense enough to jump through places that have been burnt. As anecdata, I grew up in Southern Australia and little bushfires were common every summer. Occasionally a big one. I've moved to south east Queensland and never really seen any significant bushfires in the past decade (if I did it was government doing a planned burn in a national park). It is sub tropical, usually pretty wet in summer (sometimes with a drought, but usually much, much wetter than say east coast of tasmania). These last few years are different. The national parks I go to for walks are super dry. Creeks are dried up that are usually running. I had that scary feeling I had from my youth of smelling smoke on a hike and worrying about fire. It isn't something I have had for a long time.
You need to start from what sort of accuracy do you need for the task from a business perspective (including what is acceptable in terms of false positives and false negatives). Just back of the envelope stuff. You have a rough idea of the "I copied stuff other people has done rate" and the "I spent few a days mucking about rate". This stuff always follows a logistic curve with time, starting at your first rate and asymptotically going to high 90%. Use this to get a ball park estimate of how long it will take / cost. If the accuracy required is close to 100% you can probably give up straight away. For things like this that I have done in the past, a good mental model has been if it isn't worth "manually automating" the task (i.e. paying someone somewhere to watch a webcam and send the email so you always have the end product and you eventually get labeled data as a byproduct) it might not be worth trying to automate it.
It is just "what you do". If it is a small problem the default is qr decomp of A. If you are worried about speed do a cholesky decomp of A'A. If the problem is big (usually because of a sparse A) then you do conjugate gradient (because fill in will bite with a direct method). If it is really, really big (A can't fit in memory) then it isn't clear what the "thing to do" is. It is probably "sketching" but in ML/neural network land everyone just does SGD, which you can think of as a monte carlo estimate of the gradient (A for a linear problem). Maybe sketching and SGD are equivalent (or an appemroximation). "what you do" is based on convergence and stability characteristics.
Depends, if it was a constructive proof by example for a useful problem (e.g. somebody finds a provable polynomial algorithm for the traveling salesman problem). Even without the polynomial reduction to other problems that would be pretty great...
My favorite story about researchers getting by without phds is about Lars Onsager, quoted from his Wikipedia article
At Yale, an embarrassing situation occurred: he had been hired as a postdoctoral fellow, but it was discovered that he had never received a Ph.D. While he had submitted an outline of his work in reciprocal relations to the Norwegian Institute of Technology, they had decided it was too incomplete to qualify as a doctoral dissertation. He was told that he could submit one of his published papers to the Yale faculty as a dissertation, but insisted on doing a new research project instead. His dissertation, entitled, "Solutions of the Mathieu equation of period 4 pi and certain related functions", was beyond the comprehension of the chemistry and physics faculty, and only when some members of the mathematics department, including the chairman, insisted that the work was good enough that they would grant the doctorate if the chemistry department would not, was he granted a Ph.D. in chemistry in 1935.