Federated knowledge graphs/underlay

In the world of knowledge graphs, an underlay network is a graph network connecting observations, concepts, entities, relationships and other layers in a knowledge ecosystem. The underlay project tries to identify and connect all public underlays.

Where overlays are views of knowledge, often scoped by a set of category, filter, query and other lenses, they can be seen as an interface to curated + structured layers of knowledge, themselves grounded in elements of underlay networks.

A version of this is the routing tables for physical infrastructure[1] that facilitates packet delivery in a computer network. Another is knowledge infrastructure including tools and protocols for grounding public knowledge graphs in shared concepts, vocabularies, and workflows.

Underlay network concepts edit

  • In computer networking: an underlay network is the physical network of underlay devices connected to one another, which determine connectivity and handle IP routing. These use underlay protocols to route communication across the network. This network helps maintain availability and ensure performance as the network topology changes. [2]
    • The rise of software-defined networking created tools operating as interconnection between underlay and overlay networks.[3][4] To ensure a network is interoperable with any controllers, and with a range of physical and wireless networks, the data plane used as the underlay for the network should restrict itself to using public standards.[5]
  • In spectrum sharing and IoT networks, where a secondary system is sharing spectrum with a primary one: an underlay approach assumes that the secondary system always has access to the spectrum, but has to satisfy a constraint when doing so, that limits the interference experienced by the primary system.[6]
  • In knowledge organization, a network of knowledge graphs, each with their own underlay of claims and local context, may share schemas and vocabularies, and cross-reference one another, enabling the construction of joint graphs.

Underlay protocols edit

  • In computer networking: routing protocols include ethernet switching, VLAN, and routing (OSPF, BGP, IS-IS...) IS-IS, for instance, can keep track of the local routing neighborhood and peering capabilities, agnostic to whether traffic is IP (v4 or v6) or non-IP.
  • In knowledge organization: shape and schema languages (such as ShEx, SHACL, and TASL) are used to support reuse of datasets, and interoperability across different tools for storing data.
  • Reconciliation APIs and protocols for mapping concepts in different namespaces to one another. (cf. Wikidata reconciliation via OpenRefine[7])

Related concepts edit

A universal authority file:

Overlay publishing:

  • In maps: geo overlays[8]
  • In journals: Overlay journals[9]
  • data overlays + analysis

What a knowledge underlay is not edit

Microstation underlays, carpet underlayment, or a schlock Malzberg novel.

References edit

  1. Bhardwaj, Amit (2022-05-12). "Difference between Underlay Network and Overlay Network - IP With Ease". ipwithease.com (in en-US). Retrieved 2023-03-07. 
  2. "Difference between Underlay and Overlay Networks". Route XP Private Network Services. Retrieved 2023-03-07. 
  3. "Data Center Networks | Juniper Networks US". www.juniper.net. Retrieved 2023-03-07. 
  4. Lu, Jie; Callan, Jamie. "Content-Based Peer-to-Peer Network Overlay for Full-Text Federated Search" (PDF). CMU. 
  5. Ullal, Jayshree. "The Ideal Cloud Network: SDN Overlays, Underlays or Both?". blogs.arista.com (in en-us). Retrieved 2023-03-07. 
  6. Gu, Yifan. "Minimizing Age of Information in Cognitive Radio-based IoT Systems: Underlay or Overlay?" (PDF). [[w:arxiv.org|]]. 
  7. "Reconciling with Wikibase | OpenRefine". openrefine.org. Retrieved 2023-03-07. 
  8. "Overlap Among Different Datasets - AbXtract — Orion Programming Guide 2021.1.0 documentation". docs.eyesopen.com. Retrieved 2023-03-07. 
  9. Xie, Cheng; Zeng, Ting; Xiang, Hongxin; Li, Keqin; Yang, Yun; Liu, Qing (2021-02-26). "Class Knowledge Overlay to Visual Feature Learning for Zero-Shot Image Classification". arXiv:2102.13322 [cs].