Darius Afchar

Darius Afchar is a researcher at Deezer, and a recent graduate from Sorbonne University in France, where he completed his industrial PhD in 2023 on machine learning interpretability for music recommender systems. Darius has been working in the research team at Deezer for 7 years and explored various topics, including :

- Researching feature attribution, both experimentally and theoretically to better understand how recommendations are computed and why ;

- Retrieving musical concepts from trained AI models ;

- Working out some novel theoretical results on matrix factorisation, to bridge the gap between geometrical properties of the latent space recommenders operate in, and the real social phenomena they emerge from, as the existence of music communities or popularity biases.

More recently, with the popularization of AI services as Suno and Udio, Darius Afchar has shifted to the topic of AI music detection. He and his co-authors published the first papers on AI-music detection (and the challenges with it). In turn, this has led Deezer to be the first music streaming company to provide an AI tag in its app, providing transparency to its users, and fostering a debate on the impact of AI on human creation and the protection of artists.

Beyond his published work, Darius is very interested in philosophy, responsible AI, and its governance (as with the upcoming EU's AI Act).He doesn´t believe that technology is neutral or that there is only one path forward with innovation, but that we should make it more participatory and democratic with all stakeholders involved in and impacted by these socio-technical systems.

Darius is also a piano player, and has a funk and neo-soul band. He is in love with the sound of Rhodes keyboards and with the albums of Herbie Hancock.