Leveraging Semantics for Incremental Learning in Multi-Relational Embeddings
This addresses incremental learning for service robots that need to incorporate new semantic concepts, though it appears incremental/hybrid in approach.
The paper tackles the problem of incremental learning in multi-relational embeddings for service robots, presenting Incremental Semantic Initialization (ISI) which improves immediate query performance by 41.4% and reduces epochs to convergence by 78.2% on AI2Thor and MatterPort3D datasets.
Service robots benefit from encoding information in semantically meaningful ways to enable more robust task execution. Prior work has shown multi-relational embeddings can encode semantic knowledge graphs to promote generalizability and scalability, but only within a batched learning paradigm. We present Incremental Semantic Initialization (ISI), an incremental learning approach that enables novel semantic concepts to be initialized in the embedding in relation to previously learned embeddings of semantically similar concepts. We evaluate ISI on mined AI2Thor and MatterPort3D datasets; our experiments show that on average ISI improves immediate query performance by 41.4%. Additionally, ISI methods on average reduced the number of epochs required to approach model convergence by 78.2%.