The Annual Meeting of WARA Media & Language

This year’s edition of the Annual Meeting of WARA M&L was held in Stockholm on November 21-23 at the beautiful Hasselbacken at Djurgården.

The theme of Day 1 was “Language modelling and data sharing” and Johanna Björklund kicked off the agenda with a summary of the arena’s activities in 2022 and our plans for 2023. The research focus is now being sharpened from analysis and generation in general, to data synthesis, applications to artificial human communication, and graph-based methods.

The first speaker was Magnus Sahlgren from AI Sweden on “Language modelling for the Nordic languages” and the GPT-3 modell that WARA M&L is building together with AI Sweden, RISE and NVIDIA.Then we had Salla Franzén from IKEA who talked about “Data driven value creation” and giving examples from the way IKEA is working.

Nithesh Chandher Karthikeyan from Umeå university showcased the results from WASP Summer School on Synthesis of Human Communication. He covered the method that was used for the student assignments, an ML pipeline involving text-, speech-, and gesture generation, and also demonstrated one of the student works. 

After that, Gabriel Eilertsen from Linköping University talked about “Synthetic data for data-centric ML” and after him we had Zenodia Charpy NVIDIA on “Applications of very large Language Models”. Frank Drewes from Umeå University concluded the day with an introduction to “Graph Generation”, covering traditional rule-based methods and sketching hybrid approaches that combine continuous and discrete states.

The theme for Day 2 was “Data generation and graph ML” and the program became more interactive with group discussions. First, Sandor Albrecht  and Johanna Björklund held an interactive session on “Data sharing for Media and Language applications”.

After this, Sonja Buchegger from Kungliga Tekniska högskolan talked about the important topic of “Privacy-preserving Data Synthesis” and Zenodia Charpy from NVIDIA gave a second talk on the topic of “Synthetic data generation with language models”.

Buchegger and Charpy were followed by Paris Carbone and Sonia Horchidan from Kungliga Tekniska högskolan who presented “ORB: The Missing Architecture for Graph Exploration”. Next, Anastasia Varava from SEB facilitated a session on “Graph ML ”and “Synthetic Data Generation” where we engaged the whole group to gather requirements from the community.

Day 3 was all about “Game Testing”. Konrad Tollmar from Electronic Arts (EA) was the first man out and discussed testing games using AI/ML. We also had Linus Gisslén from SEED and Sahar Asadi from King presenting on the topic.

After the presentations there were interactive discussions about methods and practices of game testing and a platform for game ML testing was discussed.

All in all, we had days packed with knowledge exchange, engaging speakers, and community interaction.