RT @kfountou: Positional Attention is accepted at ICML 2025! Thanks to all co-authors for the hard work (64 pages). If you’d like to read t….

RT @aseemrb: My PhD thesis is now available on UWspace: Thanks to my advisors @kfountou and Aukosh Jagannath for t….

RT @PetarV_93: "Energy continuously flows from being concentrated, to becoming dispersed, spread out, wasted and useless." ⚡➡️🌬️. Sharing o….

RT @kfountou: Positional Attention: Out-of-Distribution Generalization and Expressivity for Neural Algorithmic Reasoning. We propose calcul….

RT @kfountou: I wrote a blog @Medium on "Random Data and Graph Neural Networks" . Link: . I cover a range of topics….

RT @backdeluca: For those participating in the Complex Networks in Banking and Finance Workshop, I’ll be presenting our work on Local Graph….

RT @kfountou: Paper: Analysis of Corrected Graph Convolutions. We study the performance of a vanilla graph convolution from which we remove….

RT @zdeborova: Emergence in LLMs is a mystery. Emergence in physics is linked to phase transitions. We identify a phase transition between….

RT @kfountou: Alright, I have some important news (at least for me). Now there exists an accelerated personalized PageRank method which is….

RT @siam_acda: SIAM Conference on Applied and Computational Discrete Algorithms (ACDA23).May 31 – June 2, 2023.Seattle, Washington, U.S. N….

RT @siam_acda: SIAM Conference on Applied and Computational Discrete Algorithms (ACDA23), May 31 -- June 2, 2023. .

RT @kfountou: Open problem: accelerated methods for l1-regularized PageRank.

RT @kfountou: Does it matter where you place the graph convolutions (GCs) in a deep network? How much better is a deep GCN vs an MLP? When….

RT @HannesStaerk: New video with Prof. @kfountou explaining his paper "Graph Attention Retrospective" is now available!..

RT @HannesStaerk: Tomorrow's GraphML discussion will be with Prof @kfountou about his paper "Graph Attention Retrospective": .

RT @kfountou: New paper "Graph Attention Retrospective". One of the most popular type of models is graph attention networks. These models….

How to do local clustering on complex hypergraphs without losing rich higher-order relations? Visit us at #Neurips2021 and check it out!. When: Thu Dec 9, 11:30 AM - 1:00 PM (EST).Where: Slot D0 at Joint work with @kfountou @PanLi90769257

RT @konstmish: Newton’s method is a great heuristic but it doesn’t work globally (and line search doesn’t fix it). Cubic Newton works globa….

. does not look like message-passing anymore. It remains very unclear to me that how general SCMs (e.g., except the linear SCMs) could be converted to standard massage-passing GNNs. Would be very interested to find it out.

. because Theorem 1 in the paper considers a sum of only n, instead of 2^n, univariate functions. Although the additional “noise” term may be used to account for terms that are left out by the linearly many n univariate function, the resulting computational layer .