iflp

It's probably way easier to decide on a subject you want to learn more about, and find the most approachable textbook / lecture notes on e.g. StackExchange. It doesn't make sense to choose courses (or even subjects to learn) based on whether video lectures are available.

I don't maintain a collection of these, but I can probably recommend

* Stephen Abbott, Understanding Analysis, for mathematical analysis and a little bit of real analysis

* David Williams, Probability with Martingales, for advanced probability

* Dym and McKean, Fourier Series and Integrals

Alternatively if you have a more cursory interest, you can probably check out Metacademy [1] which provides roadmaps to learning a variety of math (and ML) concepts.

[1]: https://metacademy.org/browse

And for those who didn't follow the link, Shel actually clarified on this there: https://news.ycombinator.com/item?id=2977247

This is not exactly my area of expertise, but seems the author should read

Dawid, A. Philip. "The well-calibrated Bayesian." Journal of the American Statistical Association 77.379 (1982): 605-610.

which explains how you should interpret randomness in forecasting non-repeatable events. I don't know if 538 uses Bayesian modeling though. I think Andrew Gelman had written about this once, and as I just checked he just wrote another post on election forecast [1].

[1]: https://statmodeling.stat.columbia.edu/2020/07/31/thinking-a...

Vixra is not usually considered as a credible source.

> the pause on fees for online events by Apple is happening for all developers

Content reflowing isn't easy for pdf files. Xodo has this functionality for years and it rarely get the line break right, so there is some use for ML in this. (Or do you happen to know a pdf reader that is better in this?)

> I'm not convinced the same won't happen to JAX

And now there are already multiple NN libraries for JAX from Google...

You can't blame the TF people for things like while_loop. Those are inherited from Theano, and back then the dynamic graph idea wasn't obvious.

JAX is indeed a different situation as it has a more original design (although TF1 came with a huge improvement in compilation speed, so maybe there were innovations under the hood). But I don't know if I like it. The framework itself is quite neat, but last time I checked, the accompanying NN libraries had horrifying designs.

Make sure to check out the game. Best tetris ever.

https://qntm.org/files/hatetris/hatetris.html

It depends on what kind of understanding you want to achieve. It can be helpful to think of DNN as approximating the corresponding infinitely-wide versions. Depending on how you deal with certain scaling, they then act like a linear filter of the error signal in function space, or for single-hidden-layer networks at least, an interacting particle system. In both cases you can understand the convergence of gradient descent training using these analogies, although gaps from real-world practice exist.

I appreciate the effort the authors put to this post, but this is like saying DNNs are stacked logistics regression: the connection is superficial, and doesn't lead to deep insights about how they really work.

Reminds me of elvish (elv.sh), which has a similar focus.

Mathoverflow is different from most of the other SE sites in that it's only for research level questions. There is a separate site, math.stackexchange, for other math-related questions.

> Even if you follow all the best practices around emails, genuinely want to help them, and spend a long time writing a personalized email, people still don’t respond.

How do you write a personalized email for a new user?

Coming from an academic background where I need to reference theorems and equations frequently, I've found pandoc and madoko closest to usable. But plain LaTeX (with the help of macros) turns out to be more convenient.

The RFA source in the taiwan news mentioned the replacement took place in Feburary, and you can see that Zhong Nanshan remains in a leadership role up to yesterday [1]. A similar source [2] to the RFA one mentioned that Zhong was replaced on a post called "新冠病毒专家组组长" (leader of the COVID expert group); he's more commonly known as "国家卫健委高级别专家组组长" (leader of the senior expert group in national health commission; see [1]). Judging from the titles it's likely normal personnel changes.

Not to downplay the issue here: there have been reports of doctors being forced to include herbal medicines in the treatment from higher up, despite having little evidence supporting their efficacy. Just a random fact check.

Sources (all in Chinese; note epochtimes doesn't have a particularly good reputation):

[1]: https://www.fmprc.gov.cn/web/fyrbt_673021/t1755063.shtml

[2]: https://www.epochtimes.com/gb/20/3/12/n11934088.htm

Could you please provide the reference? The ventilation procedure is very difficult and dangerous to carry out, and I've seen people confusing oxygen mask and related devices (as treatment for mild syndromes) with ventilators.

It's been posted and attacted some attention:

https://news.ycombinator.com/item?id=21833718

This is fantastic. I've always found it cumbersome to write on onenote and have a browser / PDF reader open simultaneously. Split screen doesn't really work, especially in landscape mode.

Might finally be able to drop the paper notebook in my backpack if they can make screen larger on this one.

This is an interesting read, but I'd be interested in a more large-scale benchmark. Right now it could be possible that the SAT solver is just better optimized for the particular platform, instead of the SAT solver being superior algorithm-wise.

For Japanese it can be different, as most structural words are in Kana, a phonogram, instead of in Kanji.

Oops. I forgot that.

Tangentially, regarding your other comment that surveillance systems are often set up in an ad hoc way, could you provide some references? This is quite interesting to hear.

Public APs don't require ID for login. Only the IM services require ID.

Anyone know how to access the paper? My university has subscription for PNAS, but I can't open the pnas.org links found on Google.

(trying to) publish in CS conferences isn't as painful as in other disciplines, but without help from someone with an academic background, it's likely your work will not receive much attention, as you might lack paper-writing skills / knowledge of prior work.

If you can't find a local professor whom you can trust and who is willing to help (which you should try first), you might want to ask on r/MachineLearning. Many users there are grad students / researchers working in ML/AI, and I've seen someone in a similar situation (asking for endorsement to submit on arXiv) received help.

Approximately. I don't think there is anything special about his work being banned before - it's typical to ban works not aligned with the communist ideology before the reform in 1980s.

On exile: like many intellectuals, he fled mainland China before the founding of PRC.

> That is insane.

No, it's not. It's just easier to learn to entertain the average people, than to entertain the smaller group of more tasteful people, or to learn and sell things.

I was also quite impressed when I learnt the figures, and am still impressed at the fact that it managed to capture the mode of people's interest, but my every attempt to use Toutiao has ended in pain. I certainly spent longer than one day.