If, like me, you're not a real mathematician but suffered through linear algebra and differential equations, you can still totally understand this stuff! I started off teaching myself differential geometry but ultimately had far more success with lie theory from a matrix groups perspective. I highly recommend:
My friends were all putnam nerds in college and I was not, and I assumed this math was all beyond me, but once you get the linear algebra down it's great!
Is there anything like this for more commodity arm cores (neoverse v2) or do we think the insights from apple silicon cores will generalize well to those other ARM architectures?
I am somewhat rusty on my undergrad quantum, but I'm not entirely sure I agree with this analysis. Could you perhaps explain it more clearly in baa-ket notation?
> At some point my daughter, who is now 12, and is a crucial person in my life, enlightening my days with her intelligence, creativity and love, wanted to visit NYC for her birthday.
I know this wasn't the point of the post, but it was the most beautiful thing I've read all week, and really sums up how I feel about my own children. A small aside in a much longer post but incredibly humanizing and wonderful.
Author of that paper here -- What an incredibly nice comment to wake up to at 3am, thank you! This paper made me quit neuro and switch to systems that are easier to model (small molecules) which it turns out are still annoyingly difficult!
This is an incredible post, and thank you for the mission critical work that you do. I'm always curious what other people's work lives are like -- can you talk a little bit about your job, the tasks you're working on today, and how the people you interact with on a day like today (customers, clients, coworkers) could make today a little easier / more fun?
Many existing version control tools help with this, but there are some challenges that crop up when doing this with jupyter notebooks:
1. Versioning notebooks in a semantically-useful way is difficult: think of it more like versioning binary data than source code. I know we struggle with this a fair amount.
2. There are many debates, some here on HN, about "best practices" for collaborative versioning (do you squash? rebase? etc). These questions exist for notebook users too, and there's a lot of interest in developing tools that either make these decisions for you (are opinionated), integrate them more into the GUI nature of notebooks, or both.
This is OT but can you share a little bit more about what it's like living in East Idaho? I grew up in Boise and now live in Chicago but with the way things have been going lately I've been considering moving back, and there's a real appeal to living in a smaller town in the eastern part of the state (my dad was from Twin Falls). Internet access is always a bit of a concern though, which maybe Starlink ameliorates? Sorry to be so off-topic but you're the first person I've seen on HN from eastern Idaho!
I desperately hope this is the case, but a quick google couldn't find a link. Do you happen to have a citation that you'd recommend? I say this as someone who is a bit paranoid about hand cleanliness _and_ is very tired of cracked skin.
I used to build BMI systems in graduate school, from the lowest level (mixed-signal analog design for 70 uV extracelleular signals) to DSP (128 DSPs doing real time analysis) to the network (built my own ethernet MAC, foolishly!) to all the vis and RT-linux-based analysis. I left the area and switched into ML in grad school, but if I had to do it all over again there's one thing I think is missing:
Optics. Optics optics optics.
A tremendous amount of neural interfacing, especially in non-human primates and other organisms, is done via optics. ~All the advances in neural data acquisition over the past decade have been optical. Microscopy is the future for a tremendous amount of neuroscience and more and more people are considering it seriously for human-scale BMI.
I know optics isn't always thought of in an EE context, but it should be! Many people doing amazing computational imaging and optics work are in EE departments. Computational imaging is the new hotness and can let you combine your existing CS skills with signal processing and optics to do things like build a lensless camera! https://waller-lab.github.io/DiffuserCam/
If I were you I would ditch the RF part of your plan and study optics. Yeah, it's all EM, but the order-of-magnitude differences in the frequencies involved makes the underlying engineering quite different.
I don't do much neuroscience anymore but classic Golgi and Cajal drawings still move me to tears. The mystery these scientists began unlocking 100 years ago is still the biggest in all of science. For people interested in the modern versions, I highly recommend checking out Janelia's drosophila connectome data: https://elifesciences.org/articles/57443
I want to encourage people to think of sympy not just as a competitor to Mathematica but additionally as an incredibly valuable library that can be used _inside_ of other projects. Sometimes, you just want to compute an antiderivative, or you want user-supplied functions that you can manipulate easily, or you want to do some actual algebra.
Think of it less as a Mathematica replacement (like "Linux on the Desktop") and more as a crucial library enabling a lot of fun new creative things (like "embedded linux running on your toaster").
For example, in some of my computational chemistry work we use it to allow users to specify certain functionals, which we can then manipulate symbolically, do expression reduction and elimination on, and prove certain properties about. It's great!
I'm entirely willing to believe this to be the case; every paper I've written has been in LaTex and while the tooling has improved (emacs, overleaf, now even VScode's latex mode is great), there's still a learning barrier.
But one thing I want to stress is that LaTeX lets you produce _camera-ready_ documents with minimal additional effort. As we seek to reduce the cost of publishing and reduce the power and overhead of traditional journals, the legions of layout editors and horrible proprietary expensive CMSes need to be disrupted somehow.
What the hell, I just tried this and it turns out to be true. Sometimes I think I should go work for google for a year just so I could rage internally about this sort of shit. I too have a GAFYD account I've been paying for since 2009.
This is an incredibly interesting question, and the answer is an emphatic YES! Many systems involve iterative schemes, where the output of one step is used as the input to the next step. Here, these precision errors can accumulate, and if there's a multiplicative term in your equations, they can explode!
These sorts of problems are actually very common in a lot of scientific computing and simulation contexts, which is why many in the scientific computing community look aghast at the rise of FP16 (and even fp32) in machine learning applications. Of course, those algorithms are often of a _very_ different nature from (say) the large-scale linear algebra or PDE solvers we're using, but still it's pretty shocking if you're used to worrying about machine precision!
Comments like this always make me sad. Not just because I disagree with them (I do!) or that I think they are claiming certainty about facts in dispute (I think they are ! [1, 2] ) but because they make no mention of trade-offs. We can vehemently disagree about these things, get our information from different sources, and advocate for different policies, but hopefully we can all agree that there _would_ be tradeoffs? What might be the downsides to the proposed approach ? Do we think this would result in reduced innovation? In fewer potential cures?
Matt Levine has an excellent explanation of why this isn't as big of a deal as it seems (read down to Heartbeats) [1]
> On the other hand I am not convinced that one should look at the transaction in isolation here. My view of the situation is not only that “an ETF is a mutual fund that doesn’t pay taxes,” but also that everyone accepts that. There just seems to be broad agreement among investors and regulators and policymakers that an ETF is supposed to be tax-efficient, that ETF investors get to defer capital gains until they sell their shares. (Again: This is a very widely advertised benefit of ETFs. 7 ) Some ETFs ran into a bit of a technical problem that might have required them to pay taxes, and so they developed a very technical solution that fixed it. The fact that the solution is a little shammy-looking would be a problem if everyone expected them to pay the taxes, but since people don’t expect that, any old solution will do.
I actually agree with many of the suggestions in the paper, but of course am biased as these are my colleagues and we disagree on some of the interesting capabilities of modern serverless architectures (see https://arxiv.org/abs/1810.09679 ) . But it comes across like this DeWitt/Stonebraker piece from back in the day, "MapReduce: a major step backwards" https://homes.cs.washington.edu/~billhowe/mapreduce_a_major_...
https://www.amazon.com/Lie-Groups-Introduction-Graduate-Math...
and
https://bookstore.ams.org/text-13
My friends were all putnam nerds in college and I was not, and I assumed this math was all beyond me, but once you get the linear algebra down it's great!