Relational Retrieval: Leveraging Known-Novel Interactions for Generalized Category Discovery

The study introduces a novel approach for Generalized Category Discovery (GCD) called Relational Pattern Consistency (RPC), which integrates labeled and unlabeled data through bidirectional knowledge transfer. RPC utilizes One-vs-All classifiers for soft ID/OOD decomposition and enhances category discovery by leveraging invariant relationships among samples, resulting in improved relational pattern matching. This method achieves state-of-the-art performance on various benchmarks, offering practitioners a robust framework for enhancing category discovery in AI systems.