Where does network collide?

A network is a graph: a set of nodes (or vertices) joined by edges (or links), studied purely for its structure. In Networks: An Introduction Newman treats it as a mathematical object whose properties, such as degree distribution, path length, clustering, and centrality, can be measured and compared regardless of what the nodes stand for. The nodes may be atoms, web pages, or people, and the theory is indifferent, because the edge is an abstract relation rather than a wire or a friendship. What the construct names is topology, and any claim it supports is a claim about the pattern of connection and nothing more.[1]

Unit

A graph of nodes and edges, treated as an abstract structure

Recovery means

Not applicable

Measured by

Structural quantities such as degree, path length, clustering, and centrality

A social network is a structure of relationships among a set of social actors, whether people, groups, or organizations, tied to one another by acquaintance, exchange, or influence. Wasserman and Faust's Social Network Analysis formalized the study of these ties, offering measures of centrality, density, and position that describe an actor's place in the web rather than the actor's own attributes. Granovetter's account of the strength of weak ties showed that the character of a tie matters, since weak bridging ties reach information that strong clustered ties cannot. Here the nodes are human actors and the edges are observed social relations, so the object is a pattern of human relationship, not an abstract graph and not a mesh of machines.[2][3]

Unit

A set of social actors and the relational ties among them

Recovery means

Not applicable

Measured by

Tie strength, centrality, density, and structural position among actors

An artificial neural network is a layered arrangement of simple weighted units whose connection strengths are adjusted so that the whole approximates a function from inputs to outputs. Rumelhart, Hinton, and Williams gave the back-propagation procedure that tunes those weights by propagating error backward through the layers, letting the network learn an internal representation of its task. LeCun, Bengio, and Hinton later surveyed how deep stacks of such layers came to dominate perception and language tasks. The nodes here are artificial units and the edges are trainable numeric weights, so the network is a function approximator fitted to data, not a graph studied for its own structure and not a group of people.[4][5]

Unit

A trainable function approximator built from weighted units and connections

Recovery means

Not applicable

Measured by

Predictive accuracy or loss on held-out data after the weights are trained

A communication network is a set of interconnected devices that exchange information by passing packets between them across shared links. Cerf and Kahn's A Protocol for Packet Network Intercommunication set out how independent packet-switched networks could be joined into one internetwork, with gateways forwarding packets and hosts addressed end to end, the design that grew into the Internet. The nodes here are physical or logical devices, hosts and routers, and the edges are communication channels that carry traffic. What the construct names is a working system that moves data, not an abstract topology, not a web of human ties, and not a trained model.[6]

Unit

Interconnected devices exchanging packets over communication links

Recovery means

Not applicable

Measured by

Operational quantities such as throughput, latency, and packet loss