Graph autoencoders are a type of machine learning model designed to work with data that can be represented as graphs, such as networks of people or connections between items. They learn to compress the information from a graph into a smaller, more manageable form, then reconstruct the original graph from this compressed version. This process…
Subgraph matching algorithms are methods used to find if a smaller graph, called a subgraph, exists within a larger graph. They compare the structure and connections of the nodes and edges to identify matches. These algorithms are important in fields where relationships and patterns need to be found within complex networks, such as social networks,…
Temporal graph embedding is a method for converting nodes and connections in a dynamic network into numerical vectors that capture how the network changes over time. These embeddings help computers understand and analyse evolving relationships, such as friendships or transactions, as they appear and disappear. By using temporal graph embedding, it becomes easier to predict…
Graph-Based Recommendation Systems use graphs to model relationships between users, items, and other entities. In these systems, users and items are represented as nodes, and their interactions, such as likes or purchases, are shown as edges connecting them. By analysing the structure of these graphs, the system can find patterns and suggest items to users…
Graph embedding propagation is a technique used to represent nodes, edges, or entire graphs as numerical vectors while sharing information between connected nodes. This process allows the relationships and structural information of a graph to be captured in a format suitable for machine learning tasks. By propagating information through the graph, each node’s representation is…
Model-free reinforcement learning (RL) algorithms help computers learn to make decisions by trial and error, without needing a detailed model of how their environment works. Instead of predicting future outcomes, these algorithms simply try different actions and learn from the rewards or penalties they receive. This approach is useful when it is too difficult or…
Multi-agent coordination is the process where multiple independent agents, such as robots, software programs, or people, work together to achieve a shared goal or complete a task. Each agent may have its own abilities, information, or perspective, so they need to communicate, share resources, and make decisions that consider the actions of others. Good coordination…
Safe Reinforcement Learning is a field of artificial intelligence that focuses on teaching machines to make decisions while avoiding actions that could cause harm or violate safety rules. It involves designing algorithms that not only aim to achieve goals but also respect limits and prevent unsafe outcomes. This approach is important when using AI in…
Hierarchical policy learning is a method in machine learning where a complex task is divided into smaller, simpler tasks, each managed by its own policy or set of rules. These smaller policies are organised in a hierarchy, with higher-level policies deciding which lower-level policies to use at any moment. This structure helps break down difficult…
Off-policy evaluation is a technique used to estimate how well a new decision-making strategy would perform, without actually using it in practice. It relies on data collected from a different strategy, called the behaviour policy, to predict the outcomes of the new policy. This is especially valuable when testing the new strategy directly would be…