nestorD

21 ms·

211.
by nestorD 5y ago
I would say it is somewhat comparable with XLA which is the compiler backend for Jax (and was heavily inspired by Halide).
212.
by nestorD 5y ago
What they call high gloss (the name is misleading) appears to be spot gloss: https://cd1.makeplayingcards.com/images/site/card-box-stock/...
213.
by nestorD 5y ago
> A couple decades ago I made a Tarot deck. https://egypt.urnash.com/tarot/ Sadly you can’t get a copy right now, I really need to get a reprint happening. It’s got spot gloss on the cards so that’s kind of complica
214.
by nestorD 5y ago
That's a good point. A think where it fail is that some candidates are not applying to classical channels (such as latin american) so recruters are not seing them.
215.
by nestorD 5y ago
Yes, that's the "in practice not all companies act as rational capitalists" part.
216.
by nestorD 5y ago
That is great! Latin american are still suspiciously uncommon in tech so there is definitely a neglected talent pool. That illustrates one of the few good traits of capitalism: discrimination pushing you to miss on good candidates (women, p
217.
by nestorD 5y ago
To die for!
218.
by nestorD 5y ago
Very cool and sensical when you pause to think about it. While different architectures might have different components, they are also likely to reuse building blocks (ie: the first layers of convolutional neural networks are famously very s
219.
by nestorD 5y ago
If you work on a particular problem (lets say because it is your startup's object of research) you very likely are testing a large number of architectures on your dataset. With this you could train a GNN on the various models you teste
220.
by nestorD 5y ago
Computing the value would be a problem but determining which is bigger is purely a CS question (just one that cannot be solved directly with our usual floating point tool: IEEE 754 arithmetic).
221.
by nestorD 5y ago
I am seing one empty box that should get a "has been tried" mark and four that should get "are being used" marks. I don't know if they work but I know they are being used for profit.
222.
by nestorD 5y ago
Part 2 to 6 of the paper explains the elements of notation I used. The algorithm I wrote had to do with both the usual matrix product and the kronecker product (in short it lets you compute the Kronecker product of several matrices together
223.
by nestorD 5y ago
Kronecker products have tricky properties and I was struggling to implement a linear algebra algorithm using some of those properties in a way that would be both correct and efficient. Using the notations from the paper "On Kronecker P
224.
by nestorD 5y ago
I got my grand-father (who is an early-adopter type of person but not tech savy) on Linux (latest LTS Ubuntu) about three years ago. Telling him he won't have to bother about antivirus and making sure it had applications and drivers fo
225.
by nestorD 5y ago
Yes! I have long thought that GPT type of model are huge because they are forced to encode a lot of raw knowledge, giving them the ability to search for knowledge in a database solves that problem which should help making them smaller while
226.
by nestorD 5y ago
The problem is not that computers are bad are math (they are arguably perfect at doing math for the operations definition they are given), it is that we think that we are using real numbers and the associated operations when we are using IE
227.
by nestorD 5y ago
My solution to that problem is to have friends with which I love to spend time. Either due to their very different background, to the brilliance of their mind or to their deep kindness and love for all things around them. Getting there did
228.
by nestorD 5y ago
To be fair, Pearson, Spearman and Kendall's tau are the coeffisicent people use in practice. Had I been in the author's position I would have done the same: cite all the interesting developements but compare with what people actua
229.
by nestorD 5y ago
I came to say that my favorite book read this year is her "The Birthday of the World and Other Stories". Very nice to read science fiction that is about people rather than things.
230.
by nestorD 5y ago
Manish Goregaokar has a wonderful blog post on garbage collectors in Rust, the current implementations, the trade-offs and where they might be worthwhile: https://manishearth.github.io/blog/2021/04/05/a-t
231.
by nestorD 5y ago
Tooling got me from Ocaml to F#, I would recommend it if that is a criteria for you.
232.
by nestorD 5y ago
It depends. I have seen some algorithms (the example that comes to mind was a clustering) become worse solely due to numerical error. When that happens, if you are not equiped to measure the numerical error or at least trained to suspect it
233.
by nestorD 5y ago
Funny thing, the very first job offer I received on Linkedin was for a F# job (I was still a student at the time).
234.
by nestorD 5y ago
I believe all big frameworks have some work being done to make them compatible with AMD GPUs. Here is the relevant issue for JAX (support seem to be in alpha but viable): https://github.com/google/jax/issues/2
235.
by nestorD 5y ago
Julia has XLA.jl [0] which interoperate with their deep-learning stack and Elixir has NX [1] which is higher level (basically JAX but in Elixir). I would love to see someone do something like that in Rust... [0]: https://github.c
236.
by nestorD 5y ago
I have a preference for Flax, you basically get Pytorch but streamlined thanks to the Jax foundation.
237.
by nestorD 5y ago
And datastructures.
238.
by nestorD 5y ago
I have been taught that the proper way is to use an algorithm that will summarize the random forest as a close-enough decision tree. I have also been taught that it can be very misleading...
239.
by nestorD 5y ago
The Enzyme team has been doing something like that (AD at the LLVM step making it available to all LLVM languages) with Rust: - https://internals.rust-lang.org/t/automatic-differentiation-... - https://githu
240.
by nestorD 5y ago
It is in French and deals with a subfield of a subfield of magic: https://the-mystery-store.com/product/fr-billets-nestor-dee/ (out of stock at the moment but the editor tells me they will refill the stock shortly

More ›