The Deep Visual Learning (DVL) Lab at Fondazione Bruno Kessler conducts research in machine learning and computer vision, with a growing focus on multimodal and foundation models. Our work aims to develop learning methodologies that make large-scale models more adaptable, efficient and trustworthy when deployed on real-world data. We are especially interested in how vision-language and multimodal foundation models can be adapted, specialized and continually updated without costly retraining, and in making them reliable: robust to distribution shifts, fair, interpretable and safe. Recent projects include training-free and open-vocabulary recognition with vision-language models, efficient adaptation and benchmarking of large multimodal models, video understanding, and human behaviour analysis.