oliwaw

Deep learning is just a rebranding of "neural networks". When neural nets became unpopular in the 90s and early 2000s, people talked about "multilayer networks" (dropping the "neural") since it wasn't really useful to think about this approach from the neuro perspective (since it's such a cartoonish model of real neural networks anyway).

Now that very deep networks have become possible, and various graphical models and Bayesian approaches have also been folded under "deep learning" (for example, using back-propagation to learn complicated posterior distributions in variational Bayes) deep learning is not just about vanilla feedforward nets.

>If you were to run almost any kind of reasonable attempt at analyzing and valuing Tesla based on what it should maybe be worth via comparables, you'd get a figure dramatically lower than it is today. Do you compare it to a solar company? To Fiat? To GM and Ford? To Porsche? Doesn't matter, every example gets a Tesla worth far less than it is today. Public shareholders disagree however, they think it has a very bright future.

If Tesla succeeds in finishing the Gigafactory on a reasonable timeframe and delivering a large number of Model 3 cars by the end of this decade, why would it be worth "far less than it is today"?

It won't be used for the rocket but for the Mars transport spaceship.

But cost is also at issue. For cargo, it might make sense to take on 0.5-1.0% risk versus the massive investments necessary for the development of a space elevator.

And when we're talking about human spaceflight, consider the work SpaceX and Blue are putting into in-flight abort and propulsive landing - SpaceX's Crew Dragon should be much safer than the Space Shuttle, for example.

Probably the biggest barrier to space elevators is exactly what SpaceX and Blue Origin are doing - safe, reliable, and eventually (relatively) inexpensive rockets will make the large fixed costs of setting up a space elevator uneconomical.

>Space-X can't make and launch Falcon boosters fast enough now.

The bottleneck is not Falcon 9 production, at this point it's largely on the payload side. A faster turnaround would certainly help, but there's only so much optimization that can be done.

The Falcon Heavy is not a priority because (1) much less demand from customers and (2) it's a transitional system between Falcon 9 and built-for-reuse BFR/MCT, so it will have limited use.

Between the recent news of SpaceX shipping a test article of its Mars rocket to McGregor (the Raptor engine) and this large contract for material for its Mars Colonial Transporter, SpaceX's Mars plans are shaping nicely.

Hopefully Congress takes notice within the next few years, and funnels the money currently spent on the SLS rocket to Musk's Mars project.

>Science looks for a system which can exist without supernatural intervention, but the creation of time/space/matter doesn't fit within those constraints.

So our universe must fundamentally be a supernatural creation?

That's an extraordinary claim, what is your evidence? We currently have insufficient data and incomplete theories to fully describe the origins of the universe, sure. But how is this different from someone 500 years ago saying:

"Science looks for a system which can exist without supernatural intervention, but the creation of the Earth and humanity doesn't fit within those constraints."

SpaceX is planning on launching a satellite internet constellation with enough satellites to cover most of the globe.

>Not necessarily, often derivatives are analytically known in ML.

The focus here is largely on deep neural networks. In this domain, the Hessian cannot be computed and SGD (with minor variants) continues to be the golden standard.

While Matlab still probably comes ahead in image processing, surely in computer vision Python is on top? You mentioned academic research, isn't the majority of modern object recognition, segmentation, etc. done with convnets? Most popular convnet libraries are in Python.