In the age of big data, grappling with demanding processing tasks is inevitable. When it comes to handling videos, the need for efficient, high-throughput solutions becomes paramount. This project delves into the setup of a multi-GPU pipeline optimized for bulk video captioning. Leveraging the torchserve server and the captioning model from Hugging Face, this project aims to enhance performance. Code provided.

Dostaję nieznaną monetę, wykonuję nią trzy rzuty i za każdym razem wypada orzeł. Jakie jest prawdopodobieństwo, że w następnej próbie również wypadnie orzeł? Czy ta moneta w ogóle jest wyważona? Jak to wyliczyć i sprawdzić?

Diffusive models provide good results when it comes to sampling the learned data. Unfortunately, most of the out-of-the-box solutions cannot access the underlying data likelihood. In this post we discuss ways to overcome it. Code is provided.

Implementation of diffusion models look hard? Do not fret - fortunately there’s no need to introduce fancy ML techniques to understand the underlying mechanisms. In this post I discuss how to learn a 2D synthetic dataset with a simple vanilla feed-forward network (or the nano diffusion). Code is provided.

Interested in AI? Heard of Smoluchowski? Have you stumbled lately on any breathtaking AI-generated images? Why am I asking you these questions? Find out here.

What is arbitrage? Has it gone extinct in the modern age? How to spot it in the wild? Is there any physics involved?

This is the last part of exploratory work on wykop.pl started here and here. We focus on finer activity of both single voters and voter groups. Full data can be found here.

This is a continuation of the exploratory work of wykop.pl I started some time ago. This part focuses on voting strategies. Full data can be found here.

For some time I wondered about the inner workings of wykop.pl, one of Poland’s main social networks not related to GAFA behemoths. In particular, there is an ongoing discussion about the existence and character of vote brigading campaigns on the website. Full data can be found here.

This is my personal website where I write, present projects, and discuss ideas.