Polarace

@flxia_ @GraphJJA @FangYi11101 This isn’t quite right, though it’s an easy mix-up. A single sum X+Y need not be Gaussian: two dice give a triangular law, while if X,Y \sim \mathrm{Exp}(1) are i.i.d., then X+Y \sim \Gamma(2,1). The CLT is about summing many independent variables, not just two.

@GraphJJA @FangYi11101 it's just that convergence happens as we increase the # of summed RVs, so with only 2 we just get a triangle shape (which is only a rough approximation of a normal distribution) but that's still the CLT in action

A subtle point, but this isn’t how the central limit theorem works. You need to sum the random variables to approximate a normal, all he’s doing is getting a histogram of a triangular distribution.

@ninja_maths Perhaps this could be improved by showing the actual Butcher tableau? This would make understanding table itself useful for further study (Butcher series, etc), rather than just a pedagogical device. But this already much nicer than how RK is usually presented :)

Another topic I'm excited to see. The Runge-Kutta (RK4) method. This is a go-to method for solving initial value problems. Numerical methods are difficult to teach in an online environment. Our first few attempts at teaching Euler's method didn't go very well. Kids struggled. Thankfully, after a few attempts, we found a methodology that seems to work. - Ask students to master calculating one iteration only in the first lesson. Multiple iterations can be done in a separate lesson. There's no need to master absolutely everything in a single task. Splitting concepts across several lessons creates "encompassings," which improve learning efficiency and also give students more reps. - Build intuition. In the Euler topic, we have students construct a regular tangent line approximation to an ODE before formalizing it using Euler's method. In the RK4 topic, we've dropped the common k_1, k_2, ... notation and replaced it with y_1',y_2',... since the latter helps reinforce the idea that what you're actually finding is slope estimates at various intermediate points. I personally hated the k_i notation when I studied this as a student. Props to Josh (my colleague) for making a good call here. - Free kids from "index hell". Too many indices often lead to confusion. Using a tabular approach can avoid many of them. Tables also help students to create mental models of the process. - Carefully scaffold lessons. The tables provide suitable scaffolding for initial learning. Students must master this first, before we take it away.

@papers_anon I would like to see these flavour of the month optimizers demonstrated on NanoGPT vs Muon baseline. I think it’s telling they rarely are.

AdaGrad Meets Muon: Adaptive Stepsizes for Orthogonal Updates Proposes a new algorithm, AdaGO, which combines a norm-based AdaGrad-type stepsize with an orthogonalized update direction. Links below

@norpadon @doomslide It’s good but more journals, etc need to accept manuscripts in this format for it to really take off. I’ve wanted to use it before but it’s gone against submission guidelines

@doomslide Thats why typst.app exists

LaTeX is insane because it appears to be one of the most convoluted trash software ever created with the potential to gigafry your brain but is exclusively used by literal turbonormies who unironically want to like "render greek letters" and basically get oneshotted by it.

@TheExtremeMusi1 Anton Newcombe, idk if it was the alcohol that made him almost hit the guitarist over the head with a guitar or if he’s just like that. Fantastic show though.

Who’s the most intoxicated musician you’ve ever seen perform live???

@Migmag789 Check out his piece Strumming Music it’s very good :)

Which one are you?

@DrewPavlou This is an old photo Drew you should delete

@predict_addict Interesting, but with no benchmarks on OGB and no comparison with graph transformers (i.e., GPS, graphormer, etc) I wouldn’t be so certain about calling this SOTA.

KA-GNN: KOLMOGOROV-ARNOLD GRAPH NEURAL NETWORKS FOR MOLECULAR PROPERTY PREDICTION ”Molecular property prediction is a crucial task in the process of Artificial Intelligence-Driven Drug Discovery (AIDD). The challenge of developing models that surpass traditional non-neural network methods continues to be a vibrant area of research. This paper presents a novel graph neural network model-the Kolmogorov-Arnold Network (KAN)-based Graph Neural Network (KA-GNN), which incorporates Fourier series, specifically designed for molecular property prediction. This model maintains the high interpretability characteristic of KAN methods while being extremely efficient in computational resource usage, making it an ideal choice for deployment in resource-constrained environments. Tested and validated on seven public datasets, KA-GNN has shown significant improvements in property predictions over the existing state-of-the-art (SOTA) benchmarks.” #KAN

@FCelysiairl @nikicaga This is possible in iron curtain :)

@nikicaga Let me fund KMT remnants in Burma so they can launch small scale raids into Yunnan. It’ll be funny, pls paradox

Paradox is leaving trillions of dollars on the table by not making a Cold War game highly focused on internal management and politics

@Darsam12_ haven't read it, but read many of his articles and his twitter is good. would probably be worth a try

econ/leftie nerds: anyone has read/heard of this book and knows if it's worth a try or not?

@RedSpectre5 this screen is getting updated soon. glad you like the mod!

Colored charts

@AndreiBtvt @POLITICS_JEDI Do you have any info on this Strike Eagle analogue, such as a project number, etc? Quite interested 🤔.

@POLITICS_JEDI Initialy Soviet AF planned Strike Eagle analouge, but chief designer pressed to buld such strange design with armored cocpit like Shturmovik+Su-27 - completly unjustified design

#OTD The first flight of the T-10V-1, the Su-34 prototype, took place on April 13, 1990. The first flight of a serial Su-34 aircraft took place on October 12, 2006. How itshowed itself in a real war, we are now witnessing.

@commieposting3 Observation bias. Many of these states existed before camera cell phones were available: 1956 Tbilisi riots, about 150 were killed by the Soviet Army in Black January, about 20 killed in Baltic Janurary events, etc. And this is just a small selection in the USSR alone.

now that you mention it ive never seen footage of police brutality like this from any of the current or historical 'AES' states.

@BOBO783082 @spargles en.m.wikipedia.org/wiki/History_o…. apnews.com/article/4ee276… Wikipedia is thoroughly sourced

@leftwinglow @spargles I'm not sure that there is any substantial evidence for the use of chemical weapons? Even then i think that it is comparatively minor given the enemy they were fighting and the limited form such operations would have had to take for it not have been widely known.

@BOBO783082 @spargles Residue of Soman or Sarin nerve gasses were found in foliage and soil samples in 1988 independently by a UN chemical weapons expery. A separate UN report also identified VX residue in the same year. A gas cloud resembling mustard gas is also reported a few times, but unconfirmed.

@BOBO783082 @spargles What about the use of chemical weapons? Seems to sour the whole thing.

@spargles Fighting the apartheid regime in SAF and their allies is based actually, no matter what apologists have to say. There is literally a SAF armored car in that video from Black Ops 2.

That is all human beings ask for, that the lights be kept dim.

@BOBO783082 @maleevabraun so true

@maleevabraun Would agree except Oceania/Australia should still be a continent (because it obviously is)

@Darsam12_ @lxglb Darsam has not read Hegel

@lxglb he's overrated, seen as the guy who "exported" the revolution when many of "his things" (like the Civil Code) were just stolen from the Revolution He re-established slavery in 1802, his wars ruined France and killed millions of people in Europe. so net negative I'd say yea