Leif-Nissen Lundbæk

Amit Chaulwar, Lukas Malik, Maciej Krajewski et al.

Modern search systems use several large ranker models with transformer architectures. These models require large computational resources and are not suitable for usage on devices with limited computational resources. Knowledge distillation is a popular compression technique that can reduce the resource needs of such models, where a large teacher model transfers knowledge to a small student model. To drastically reduce memory requirements and energy consumption, we propose two extensions for a popular sentence-transformer distillation procedure: generation of an optimal size vocabulary and dimensionality reduction of the embedding dimension of teachers prior to distillation. We evaluate these extensions on two different types of ranker models. This results in extremely compressed student models whose analysis on a test dataset shows the significance and utility of our proposed extensions.

Kwok Cheung, Michael Huth, Laurence Kirk et al.

We are living in an age in which digitization will connect more and more physical assets with IT systems and where IoT endpoints will generate a wealth of valuable data. Companies, individual users, and organizations alike therefore have the need to control their own physical or non-physical assets and data sources. At the same time, they recognize the need for, and opportunity to, share access to such data and digitized physical assets. This paper sets out our technology vision for such sharing ecosystems, reports initial work in that direction, identifies challenges for realizing this vision, and seeks feedback and collaboration from the academic access-control community in that R\&D space.