jellyksong

Our team is in a similar position right now — I’m curious how you decided on Highcharts as opposed to other options like Plotly.

Basic question: Why is this faster than running Intel Linux apps in an emulated Intel Linux VM? Because Rosetta is faster than QEMU, and you only need to emulate one application rather than the entire kernel?

There's an amazing game, 4D Toys [1], that lets you interactively play with 4D objects in a 3D environment. It's great on iPad.

I believe the developer had to create a custom physics engine to support 4D space [2].

[1] https://4dtoys.com/

[2] https://marctenbosch.com/news/2017/06/4d-toys-a-box-of-four-...

I believe some algorithms, like AutoML-Zero, search on the level of mathematical operations.

This is basically what the field of neural architecture search tries to do. Here’s a good (somewhat technical) introduction: https://lilianweng.github.io/lil-log/2020/08/06/neural-archi...

In the same theme, there’s a recent documentary called “The Weight of Gold”, narrated by Michael Phelps and other Olympic champions. Really poignant stories.

> Thankfully these days Cloud Native Python apps are built around less monolithic options such as Flask, Pyramid, or Tornado.

FastAPI is also a very popular choice these days, arguably more so than Pyramid and Tornado.

Pretty cool! If you're into these kinds of historical illustrations, I highly recommend the wonderful EdX course "Visualizing Japan".

https://www.edx.org/course/visualizing-japan-1850s-1930s-wes...

Two questions:

1. I wonder how many times the test set can be used on "incremental changes in future versions of the model" before losing statistical validity.

2. This article describes their process, but not the FDA's process. Are there specific regulatory requirements for ML models beyond their four types of reports?

They show up in roots of polynomials with real coefficients.

This article is about optimization (finding good parameters), not the approximation power of neural networks (which is well-known through the universal approximation theorem).

Why doesn't this work for new industries?

Can you explain why "the data generating process of the real world is totally not obivious" is a problem for Bayesians but not frequentists?

I was thinking MPC might work for set intersection. For oblivious query, can't you do it by sending the query directly into the remote trusted execution environment, encrypted with the TEE's public key?

The article claims that oblivious query, set intersection, and machine learning on private data are not possible without FHE. However, aren't they all possible either with secure MPC or hardware based enclaves e.g. AMD SEV?

A bunch of the papers I found from a quick Google search use secure MPC rather than FHE for interbank fraud. Which one is more practical these days wrt latency?

It doesn't look like HElib implements TFHE, though. Are BGV and CKKS equally fast?

What's the advantage of using FHE in this example as opposed to Secure MPC?

Author here:

There are two ways to implement autograd, reverse-mode and forward-mode. Reverse-mode is what minigrad uses, and what most ML libraries these days use by default, since it computes gradients of all inputs (wrt one output) in a single pass. It's exactly what you describe in the 2nd paragraph.

Forward-mode autograd is the technique that can use dual numbers. It computes all gradients of one input (wrt all outputs) in a single pass. Dual numbers is a pretty neat mathematical trick, but I'm not aware of anyone that actually uses them to compute gradients.

The most approachable explanation of dual numbers I've seen is in Aurelien Geron's book Hands-On Machine Learning (Appendix D). There are articles online but I found them more technical.

Thanks for checking out the project!

The sublime.

https://en.m.wikipedia.org/wiki/Sublime_(philosophy)

How does the default NLP tokenizer work with transfer learning? Are all the fast.ai pretrained NLP models pretrained with the same tokenization scheme?

It states in the deprecation policy that "API-breaking changes will be made only in major releases (except for experimental features)".

Their career guide does specifically discuss social impact outside of your job, in terms of advocacy (what you've described), support roles, and donations.

What is the current state of the art for multi-agent DRL?

The YC company based in Shanghai is https://www.strikingly.com.

http://fetching.io does something similar.

Down in Abu Dhabi, UAE.

I played varsity lacrosse and track :)

This issue is pretty close to my heart because I'm a high school senior this year. I'm also Asian, play an instrument, have perfect SAT scores, and applied for a STEM major (well, half STEM).

I was also rejected/waitlisted from every one of my top choices.

To be honest, I'm not really sure now to respond to this article. From the point of view of a student applying to college, these "discriminations" exist pretty universally, and not just to Asians. The notion that race, gender, wealth, etc play a role in college decisions is very widespread among my classmates and peers at other schools. I don't view it as particularly bad, though, even if these colleges will never publicly admit such a thing.

College is not a completely merit based system. I think that we all like to believe that it is, but colleges also have duties outside of admitting the best students. They have to keep their alumni happy, somehow obtain outside funding, keep its population diverse enough (this is ambiguous and controversial, but I think it's a legitimate concern). Most importantly, a college also has the duty to improve society, and that's where a system like Affirmative Action comes into play. It's not perfect, but it does allow for a great deal of social mobility where society would otherwise resemble something like plutocracy. After all these considerations, there's only a fraction of each year's class that they have for purely "merit" based acceptances. And at the level that these elite colleges are at, it's almost impossible to differentiate between candidates. So this small group of accepted students is essentially random (or I like to tell myself).

There was a quote by a Yale admissions officer who lamented that he could've filled 3 identical classes with students on the waitlist.

Perhaps I'm naive, but I think that one of the benefits of our society, especially in the entrepreneurial community, is that success is ultimately the result of hard work and talent. College is not the end-all and certainly not the determinant of your future.

Changed, thanks!