dr hab. Paweł Przybyłowicz, prof. AGH, MBA’s Post

Check out my third paper 'n-dependent continuous theories and hyperdefinable sets' on arXiv! ¡Mi tercer artículo "n-dependent continuous theories and hyperdefinable sets" ya está en arXiv!

Sharing an insane paper :- Generative Powers of Ten " A method that uses a text-to-image model to generate consistent content across multiple image scales, enabling extreme semantic zooms into a scene, e.g., ranging from a wide-angle landscape view of a forest to a macro shot of an insect sitting on one of the tree branches " Paper:- https://lnkd.in/gc5bTkHK Blog :- https://lnkd.in/gystGx4j

Last week, I was in Vienna at ICLR 2024, presenting my research: What Makes a Good Prune? Maximal Unstructured Pruning for Maximal Cosine Similarity. If you missed it, the TLDR is: We show a data-free method for telling you how much to unstructured prune so that you can regain accuracy with one epoch of fine-tuning with principled reasons for why! The paper can be found here: https://lnkd.in/dcXYnubb along with a notebook demonstrating the method: https://lnkd.in/dAUmaJh7

I'm excited to share that our paper "Outlier-Insensitive Kalman Filtering: Theory and Applications" is now available on arXiv: https://lnkd.in/dhyj5eta 🎯 This work is part of my research and could not have happened without the dedicated and professional guidance of Guy Revach, Ph.D. Candidate, Nir Shlezinger, and Itzik Klein. 💡 In this work, we propose a parameter-free algorithm which mitigates the harmful effect of outliers while requiring only a short iterative process of the standard update step of the Kalman filter. Based on Bayesian learning concepts, we model each potential outlier as a normal process with unknown variance and apply online estimation through either expectation maximization or alternating maximization algorithms.

I am thrilled to share the latest publication from our research, in collaboration with Esben Taarning, Hanne Falsig, Yunfei Bai, @Juan Salvador from Topsoe shedding light on the profound influence of tin incorporation into the Beta zeolite framework. Our study combines the power of Powder X-ray Diffraction (PXRD) with Density Functional Theory (DFT) analyses to uncover intriguing insights into the structural framework of Beta zeotypes. https://lnkd.in/d4N9abN3

This paper presents a universal Cross-Attention Transformer (CAT) module for accurate and efficient semantic similarity comparison in one-shot object detection. CAT: A Simple yet Effective Cross-Attention Transformer for One-Shot Object Detection https://lnkd.in/gAWhynyw Lingqiao Liu

Get a “bird’s eye” view on GNNs in this talk with Petar Veličković. Following a quick motivation on the utility of graph representation learning, you will derive GNNs from first principles of permutation invariance and https://hubs.li/Q025Dvm40 will discuss how we can build GNNs that are not strictly reliant on the input graph structure. Last chance–last tickets: https://hubs.li/Q025DwFq0

New publication 'Latin hypercubes realizing integer partitions' by Diane Donovan, Tara Kemp, and James Lefevre. Abstract For an integer partition h1+⋯+hn=N, a 2-realization of this partition is a latin square of order N with disjoint subsquares of orders h1,…,hn. The existence of 2-realizations is a partially solved problem posed by Fuchs. In this paper, we extend Fuchs' problem to m-ary quasigroups, or, equivalently, latin hypercubes. We construct latin cubes for some partitions with at most two distinct parts and highlight how the new problem is related to the original. https://lnkd.in/gbdFxki8

New snippet (for science) just dropped. 🔥 Use this bad boy to display your calculation results in scientific notation. That means BIG numbers like `176,021,565,000` get simplified to much smaller numbers like `1.760e11`. Neat! Download: https://buff.ly/3POdNK4

What a tiny, cute, but meaningful milestone! XD In my first journal paper, ERASOR, I insisted that static map building should also be quantitatively evaluated at the voxel level. Furthermore, it is the first paper that utilizes SemanticKITTI, which provides point-wise labels for the egocentric perception, proposed by Jens Behley, my esteemed mentor, to evaluate map-centric static map building, suggesting new metrics, preservation rate and rejection rate. Furthermore, you can read ERASOR2, which accounts for instance-level information for more precise static map building while preserving most static points. I worked as a visiting scholar in Professor Cyrill Stachniss' group, collaborating with outstanding researchers Lucas Nunes, Benedikt Mersch, and Xieyuanli Chen. [ERASOR] arXiv: https://lnkd.in/gxncTc5v Code: https://lnkd.in/g4C24g3e [ERASOR2] arXiv: https://lnkd.in/gJbmUYWx Code (I'm sorry for being lazy, but we are on extending our approach now! Codes will be made available): https://lnkd.in/g-jiDUSt