Networked data management

The Networked Data Management group comprises a dedicated team of researchers specializing in various fields, including software verification, representation learning, distributed algorithms, and large-scale data management. They recognize the increasing prevalence of data with underlying graph or manifold structures, such as social networks, sensor networks, and knowledge graphs. To accommodate these relational and non-Euclidean structures, they develop new optimization methods and neural network architectures and address critical challenges related to security, privacy, and efficiency.

The group’s research spans multiple domains, including privacy-preserving machine learning algorithms, explainable representation learning, distributed algorithms for learning from data across machines, and large-scale data management for structured and semi-structured data. The treatment of relational or graphical data is becoming more crucial as it becomes ubiquitous in various domains such as social networks, biological networks, transportation networks, and energy grids. The group emphasizes the need for novel methods to construct and process suitable representations of data points in these networked contexts. Efficiently utilizing communication, storage, and computing resources across multiple machines is another key aspect of the group’s research. They tackle the challenge of learning from distributed data across multiple network locations, addressing scenarios where data is too large to fit on a single machine or constrained by privacy concerns within administrative boundaries. This necessitates the development of distributed algorithms and potentially new network architectures. Uncertain data is another significant aspect of their research. They recognize that input data in artificial intelligence systems often contain imprecision, incompleteness, or contradictions, while automatic tools used on such data produce imperfect annotations or predictions. Managing uncertainty and tracking the confidence and provenance in individual data items throughout complex processes is a major challenge. The group strives to design new programming language abstractions, verification tools, and cryptographic protocols that can provide formal verification and built-in security and privacy guarantees for AI applications.

The Networked Data Management group brings together a diverse set of researchers to tackle the challenges of security, privacy, scalability, and explainability in machine learning. Their work spans various scientific fields, with the ultimate goal of advancing the understanding and utilization of networked data in a wide range of domains.