The third summer school on Deep Learning and Large Language Models (LLM) for Natural Language Processing (NLP) was held from September 3 to 5, 2025, in Varna, Bulgaria, as part of the 15th international conference RANLP 2025.
The summer school featured morning lectures focused on theoretical foundations, followed by afternoon lab sessions dedicated to practical applications and experiments. The summer school included lectures by leading researchers in the field of NLP and Deep Learning from academia and industry.
As part of the school, a team competition was held with randomly assigned teams. The competition consisted of two tasks: Task A and Task B. Task A involved detecting offensive language at the sentence level in five different languages: English, German, Greek, Portuguese, and Sinhala. Task B was again to detect profanity, but at the word level in three different languages: English, Sinhala, and Portuguese. The goal was to train models and develop methods that would lead to the highest accuracy on a test set. Participants used the codabench platform to submit their results.
Our colleagues Nikolay Paev, Stefan Marinov, and Teodor Valchev took part in the competition. The team that Nikolay and Teodor were part of won first place in Task B.
On September 6 and 7, four half-day lessons were held on natural language processing and large language models and their development.
