eigenman

Not only is it too low, but it is the discontinuity in support. This leads to perverse incentives where for the wage you could make, you’d barely clear the threshold to lose benefits and make barely more money than on government support.

This isn’t an argument to remove support—it’s an argument to remove binary thresholds with smooth ramps to incentivize people to work.

Things may get more expensive, but if more Americans can live a middle class life even accounting for the inflation of consumer goods I think that is a good tradeoff.

There is a joke in applied mathematics that we’re like Taco Bell. We all use the same six ingredients, mixing them in different ways.

For myself, I’ve found several techniques I use over and over again. Some of this is a “when you’re a hammer, everything looks like a nail.” But fundamentally there are only a handful of ideas. One professor of mine once said the only groundbreaking result in the past few decades was compressive sensing.

If you want to start cutting money from the budget, cut the things that are the: defense spending. There is just so much waste defense procurement.

“Boss makes a dollar, I make a dime. That’s why I shit on company time.”

I concur. As a postdoc for many years adjacent to this work, I was similarly unimpressed.

The best part about PINNs is that since there are so many parameters to tune, you can get several papers out of the same problem. Then these researchers get more publications, hence better job prospects, and go on to promote PINNs even more. Eventually they’ll move on, but not before having sucked the air out of more promising research directions.

—a jaded academic

I taught numerical linear algebra in grad school and was really frustrated that even the applied math department took so long to build up to solving linear systems and eigen-decompsotions. The ordering of the material in the textbook is great, focusing on algorithms and decompositions.

“Once you have their money, you never give it back.” First Rule of Acquisition.

The first space shuttle prototype (Enterprise) started construction in 1974. The first shuttle launched in 1981. To the best of my knowledge, there were no major upgrades to the design over its career, save avionics. So even though the space shuttle was “serious space development,” it’s been a long time since a new human rated vehicle has been designed.

It would be great if that tool existed, but it doesn’t seem to right now. I can appreciate the instinct to improve packaging, but from an occasional Python developer’s perspective things are getting worse. I published a few packages before the pandemic that had compiled extensions. I tried to do the same at my new job and got so lost in the new tools, I eventually just gave up.

One of Python’s great strengths is the belief there should be one, obvious right way to things. This lack of unity in the packing environment is ruining my zen.

As someone who works in numerical optimization, this is a dirty little secret of our profession. The optimization algorithms in the literature are great at finding local minima, but often are very sensitive to the initialization as to how small the objective is. Good heuristics for initialization are thus critical for finding a good (small objective) minimizer. Sometimes this gets to the point where the local optimization algorithm does a trivial refinement of the heuristic’s solution.

Once I was working on a government funded small business grant trying to do something that was mathematically impossible (and literally the first example of intractability in textbooks of the field). The only goal was for the company to collect overhead.

(Queue Rick and Morty butter getting robot meme.)

What is my purpose?

To collect overhead.

Oh my god.

Many of the algorithms in BLAS are not easily parallelized. For example, a QR factorization an inherently sequential algorithm. Optimizing BLAS performance comes mainly from rewriting the sequential algorithm into larger blocks so as to efficiently access memory. As Jim Demmel is fond of saying, floating point optimizations are cheap, memory movement is expensive.