Owen Macindoe

Graph Comparison Using Fine Structure Analysis

We introduce a novel technique for comparing graphs using the structure of their subgraphs, which we call a graph’s fine structure. Our technique compares graphs using the earth mover’s distance between the distributions of summarizing features of their constituent subgraphs. We demonstrate the use of this technique as an abstraction of graph edit-distance and show its use in hierarchical clustering on several graphs derived from a variety of sources including social interaction data.

Comparing networks using their fine structure

We introduce a novel technique for characterising networks using the structure of their sub-graphs, which we call the network’s fine structure. To judge the similarities between networks we use the earth mover’s distance between the distributions of features of their constituent sub-graphs. This technique is an abstraction of graph edit-distance. Given these similarity measures we explore their use in hierarchical clustering on several networks derived from a variety of sources including social interaction data.

Intrinsically motivated intelligent sensed environments

Intelligent rooms comprise hardware devices that support human activities in a room and software that has some level of control over the devices. “Intelligent” implies that the room is considered to behave in an intelligent manner or includes some aspect of artificial intelligence in its implementation. The focus of this paper is intelligent sensed environments, including rooms or interactive spaces that display adaptive behaviour through learning and motivation. We present motivated agent models that incorporate machine learning in which the motivation component eliminates the need for a benevolent teacher to prepare problem specific reward functions or training examples. Our model of motivation is based on concepts of “curiosity”, “novelty” and “interest”. We explore the potential for this model through the implementation of a curious place.

Decomposing Social Networks

Networks having several hundred or more nodes and significant edge probabilities are extremely difficult to visualize. They typically appear as dense clumps, with the various subcomponents completely obscure. We illustrate a method for decomposing a network into aggregates of subgraphs whose topologies are represented as colors in RGB space.

Intrinsically motivated intelligent rooms

Intelligent rooms are responsive environments in which human activities are monitored and responses are generated to facilitate these activities. Research and development on intelligent rooms currently focuses on the integration of multiple sensor devices with pre-programmed responses to specific triggers. Developments in intelligent agents towards intrinsically motivated learning agents can be integrated with the concept of an intelligent room. The resulting model focuses developments in intelligent rooms on a characteristic reasoning process that uses motivation to guide action and learning. Using a motivated learning agent model as the basis for an intelligent room opens up the possibility of intelligent environments being able to adapt both to people’s changing usage patterns and to the addition of new capabilities, via the addition of new sensors and effectors, with relatively little need for reconfiguration by humans.