Sabitlenmiş Tweet
An interlude about structure computation, SSM and attention. Myosotis: arxiv.org/abs/2509.20503
I hope to tell more in this line of work later. See poster at SPIGM workshop.
English
Evgenii Egorov
1.8K posts
@martinmbauer If it was, than I guess we will have nice generalization of higher order svd, but we don’t.
English
Everything is just matrix multiplication. And if it isn’t it’s infinite matrix multiplication
nibbleton 🛠️@nibbletonbuilds
i just had the realisation that multiplication of complex numbers is basically just matrix muplication.
English
Inspired with arxiv discussions I propose to life ban when there are equations in paper which doesn’t match free indexes on the left and right parts of equations
English
It's hard to believe, but there was a time at which the kernel trick was considered the absolute holy grail of AI, as it made it possible to use convex optimization instead of back-prop and SGD. Incidentally, self-attention and linear RNNs feel weirdly reminiscent of those vibes.
Avi Chawla@_avichawla
Why is the Kernel Trick called a "trick"? (it's a popular ML interview question) Many ML algorithms use kernels for robust modeling, like SVM, KernelPCA, etc. The core objective of a kernel function is to compute dot products in some other feature space (mostly high-dimensional) without projecting the vectors to that space. But how does that even happen? Consider the image below. Let’s assume the following polynomial kernel function: - k(X, Y) = (1+XᵀY)². Also, for simplicity, let’s say both X and Y are two-dimensional vectors: - X = (x1, x2) - Y = (y1, y2) As shown in the image below, simplifying the kernel expression produces a dot product between the two 6-dimensional vectors. This shows that the kernel function we chose earlier computes the dot product in a 6-dimensional space without explicitly visiting that space. And that is the primary reason why we also call it the “kernel trick.” More specifically, it’s framed as a “trick” since it allows us to operate in high-dimensional spaces without explicitly computing the coordinates of the data in that space. RBF kernel is even better in this respect. It lets you compute the dot product in an infinite-dimensional space without explicitly visiting that space. I have shared an article in the comments with a mathematical explanation of the RBF Kernel. ____ Find me → @_avichawla Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.
English
O! Drop toilets! But feeling is not the same if there is not 7 them in a row.
Sara Hooker@sarahookr
I grew up in Mozambique where there are a lot of drop toilets. But typically they are always phased out in modern bathrooms. I thought it was kind of interesting Singapore has a dedicated drop toilet in its airport alongside other western style toilets.
English
There are general questions of economics which we are faced constantly but still there is no answer or this answer is not accepted by society?. For example, is it possible to build a system which rank people by credibility but doesn’t build gate keeping and if so what is system? I am sure there is a mathematical answer. Then instead of discussing is it good or bad measure we can just if it aligns with it. Or give up and say okay this is subjective do whatever you want.
English
Why people in all this linear attention paper doesn’t use einstein notation? What driven this desire to transpose everything here and there just to keep matrix, when anyway operator properties not used ( except of Myosotis, best paper in the world according to myself)
English
I need to present UMA paper recently arxiv.org/abs/2506.23971 and while there is no doubt that amount of work is non trivial, there is no details of how exactly averaging of FC layers was done in used architecture. For this I was need to go back to eSCN paper and a bit of imagination. I left in a mixed feelings
English
@DorinCheptea @alec_helbling Idk to what you answer, but I was refer that either harmonic part is missing or manifold.
English
@eeevgen @alec_helbling Pointwise addition of vector fields. It is very simple.
English
The Helmholtz decomposition is one of the fundamental results of vector calculus.
It says any well-behaved vector field can be split into two parts, one capturing sources and sinks through divergence, and one capturing rotation through curl.
English
Typical day. And my building is quite fancy, I see all this fucking posh people in it.
English
I spend like 30 minutes each day putting trash inside the trash bin otherwise it distributes across the whole street. One thing I can not understand why bother to bring trash near bin but don't put inside? Anyway, in Russia I lived in a house where were heroin dealers and guess what I didn’t have such a problem.
English
@dmitry_grinko Anyway prestige is relative to a close group of people, so inside of academia I don’t worry, they will find a way. But to justify for society why academic work is hard, worth of finding etc — this is quite unclear 😶🌫️
English
@dmitry_grinko We will bet something important, like uploading pdf to arxiv will require drop of first author blood.
English
Academic careers are mostly built on the idea of prestige. Solve an old important problem, build a powerful theory, design a new clever experiment, and you earn prestige. It is quite clear now that this currency will collapse. What will replace it, if anything?
English
Why one need equivariance if there is locality 🤔
Mohammed AlQuraishi@MoAlQuraishi
Equivariance is dead😢 Or is it?😈 Genie 3 is out! Our latest protein design model yields SoTA results for binder design & motif scaffolding, greatly improving on BindCraft & Proteina-Complexa It does so using all-atom SE(3)-equivariance on a branched polymer representation👇
English
@Kaju_Nut I hate the fact that time is in horizontal direction here... but great one...
English
@SamuelGWalters Interestingly I have a vice versa thought recently: can I extend definition of function if exists such rotation etc
English
