IJCAI, ICCV, LongLLaMA. Achievements of our researchers

IJCAI 2023

A team of scientists from IDEAS NCBR researching computer vision was admitted to the IJCAI International Joint Conferences on Artificial Intelligence Organization in Macau with the paper “Active Visual Exploration Based on Attention-Map Entropy”. The team consists of Adam Pardyl (IDEAS NCBR/Jagiellonian University in Krakow, UJ), Grzegorz Rypeść (IDEAS NCBR/Warsaw University of Technology, PW), Grzegorz Kurzejamski (IDEAS NCBR), Bartosz Zieliński (research team leader, IDEAS NCBR/Jagiellonian University in Krakow), Tomasz Trzcinski (research group leader, IDEAS NCBR/UJ/PW/Tooploox). You can see Tomasz, Grzegorz K., Adam and Grzegorz R. in the picture above.

The IJCAI conference will be held from August 19-25, 2023 in Macau.

arXiv: https://arxiv.org/abs/2303.06457

GitHub: https://github.com/apardyl/AME

ICCV 2023

“ICICLE: Interpretable Class Incremental Continual Learning” is the work of the team: Dawid Rymarczyk (Ardigen), Joost van de Weijer (CVC UAB), Bartosz Zieliński (IDEAS NCBR), Bartlomiej Twardowski (CVC UAB, IDEAS NCBR). The publication was accepted to the International Conference on Computer Vision, which will be held in Paris on October 2-6.

arXiv: https://arxiv.org/abs/2303.07811

LongLLaMA

IDEAS NCBR researchers Piotr Miłoś, Szymon Tworkowski, Konrad Staniszewski and Mikołaj Pacek as well as Yuhuai (Tony) Wu and Henryk Michalewski from Google DeepMind developed LongLLaMA – a language model based on OpenLLaMA published by Meta.

LongLLaMA is able to receive 256,000 tokens or even more, compared to OpenLLaMA, which could accept up to 8,000 tokens.

More on arXiv: https://arxiv.org/abs//2307.03170

GitHub: https://github.com/CStanKonrad/long_llama