Marta Moscati

The challenge provides a benchmark dataset, pretrained baseline models, and an evaluation framework to advance face--voice association, highlighting the need to foster the development of models that capture identity-specific aspects beyond language and gender.

Swapping cold items during training can dramatically boost recommendation accuracy and enhance user engagement with previously overlooked items.

Real-world speaker identification can thrive even with missing modalities and multilingual contexts, challenging the status quo of multimodal systems.

Autoguidance—using a model to guide itself—can effectively reduce popularity bias in diffusion recommenders, leading to fairer item exposure without significantly sacrificing accuracy.