Feasibility study of using fractal-aware features in classifying Alzheimer's disease. Features are extracted based on MRI scans while we investigate three types of classifiers:

1. k-nearest neighbor
2. support vector classifier
3. gradient-boosted decision trees

A tool for meme scraping and AI-based analysis of meme clustering. Works as a API server based on docker and written in Flask. mmx has 4 containers:

• mmx:scrape - handles online scraping of memes and saves it to the mongodb database
• mmx:feat_extract - continuously takes in the memes, calculates for each meme image a feature vector using an efficientnet-v2 model compressed with tensorflow_lite package.
• mmx:api - API server for user requests
• mmx:nginx - reverse-proxy nginx server used in a production ready API server

A tool for creating video captions in bulk. Uses the BLIP-2 + OPT img2txt model as the captioning module. Comes in two variants:

• single GPU - uses flask as the server
• multiple GPUs - uses torchserve as the server

A set of scripts for a large dataset scraping project. Used to download extensive information from the social medium wykop.pl website, available on Kaggle as a public dataset wykop-data-2022. Dataset consists of a 6 months of links published on the page with the 500 most popular tags. Each link has information on its creator, popularity and status. Used in preparation for a set of blogposts:

• What is happening on wykop? pt.1: Authors
• What is happening on wykop? pt.2: Voting Strategies
• What is happening on wykop? pt.3: Voter groups

• time-resolved activity of some users point towards bot-like actviity
• a simple criterion for link promotion was identified (see the plot above)
• existence of cooperative user clusters which upvote/downvote same links