thearn4

The Software Engineering Daily podcast has a good ongoing series on ad fraud that explores why that isn't happening.

If I understand it correctly, it sounds like they used a NN to fit a surrogate model to the kind of analytical physics-based pipeline that they had been using before?

It's an interesting alternative to visualization of tree-based hierarchies

Yeah I was hoping for at least a bit more on the Gman's motivations.

It's a good and short read, and I accept it as closure for what I always thought was a very well written series.

I think the issue is that by the time you are on descent, the flight is nearing an end, so the benefits for range extension aren't great.

At NASA we tend to have 2-3 person offices. But it's a big agency so YMMV.

Is there a biological analog of gear ratios? A nice thing about mechanical gearing is the ability to trade off speed and torque on the fly.

While this is true, the "flag planting" nature of pre-prints also kind of bugs me. But I know this has been discussed on HN before.

It was the first thought that came to me too. I was lucky enough as a PhD student (mathematics) that I never would have had any doubt that my advisor would have jumped in and fought on my (really 'our') behalf on something like this.

I feel for the author, it sounds like a terrible situation.

I remember using GAs to tune SVM kernel hyperparameters a few years ago. They worked well enough, but I mostly did it because I was lazy.

I would say the collatz conjecture is up there. It's been an entry point for many people into the computational side of mathematics.

Take any positive whole number, n.

If it is even, divide it by 2. (n = 2n)

If it odd, multiply it by 3 and add one. (n = 3n + 1)

Repeat. If n = 1, stop.

Does this process always lead to 1? It seems to, but who knows.

https://en.wikipedia.org/wiki/Collatz_conjecture

I remember being fascinated by GAs as an undergraduate, but haven't seen much discussion come out of the space in a while. I'm guessing machine learning took off (in particular deep learning) and put an emphasis on the impact of large-scale gradient-based optimization via backpropagation/adjoint approaches.

In other words, rather than apply GAs to the optimization of a large complex system, these days the paradigm seems to be to find a way to derive the needed gradients (or extract them via automatic differentiation). Does this seem to be the case, or are GAs still shining in some applications?

Casper mattresses and meundies will have to step up their game.

Still using a 2011 MBA. I just had to change out the magsafe charging port/IO board, but other than that its held up great.

Which really makes me hope that the Uber Elevate concept is smoke and mirrors, and they aren't going to try and disrupt their way through aviation regulations.

It's super retro, but part of me kind of likes the NASA "worm" logo over the "meatball".

https://en.wikipedia.org/wiki/NASA_insignia

I agree that the focus on specific tooling was tricky. That said, I really do like the notebook environment. It's mostly the AWS setup that I found a bit fragile.

4, 5, and 6 are good advice for students learning to give academic presentations as well.

i can't help but love KSP. But I wish it was easier to just design the craft and missions, and not have to do the manual control. I like flight sims, but KSP controls are a bit too wonky for my tastes.

Python3 + numpy + scipy + (possibly) tensorflow, for technical computing. The ecosystem is stable, readable, and performant enough for my needs.

As interesting as I find the current state of deep learning to be, there is something about random forests that I can't help but find much more cool. Probably the amazing out-of-box performance.

A turbine can extract energy from the expansion of any hot gas. I suppose steam is common because of how easy it is to access liquid water and concentrate it around your heat source.

Interesting, I'll have to look deeper at what has been done in this space. As I understand it, saddle points are handled by SGD-based methods because they are a bit more "noisy" in their steps through the problem space. In short, are the Krylov methods (or perhaps other approaches) a bit too precise in their attraction to critical points?

Rather than settle football rivalries, I kind of wish Cleveland would just ditch the Browns and offload them to another city already. But for good this time. Their stadium is occupying probably one of the best spots on the Lakefront, and is entirely wasted there.

The sibling comments are great, so I'll comment in a more direct application-oriented way: Krylov methods provide a way to expose most the machinery of linear algebra to you by means of only forward matrix-multiplication with the relevant matrix.

A cool thing is that means you can do linear algebra operations with a matrix (including solving linear systems) without ever needing to actually explicitly construct said matrix. We just need a recipe to describe the action of multiplication with a matrix in order to do more advanced things with it. This is why they are so popular for sparse matrices, but their suitability can go beyond just that.

Projecting problems into Krylov spaces also tends to have a noticeable regularizing effect.

Very cool! I seem to have missed that in thumbing through the outline.

It's an interesting point that robust autonomous systems may be more likely to emerge in organizations from places that have distinct seasons (including a perennial 'orange construction barrel' season).

Cleveland Clinic and NASA could have a role too, but could really use some coordination.

As a Cleveland native, I look at Pittsburgh somewhat jealously as a city that is achieving what Cleveland has been slow to do over the last 30 years: find a post-industrial midwestern identity.