Towards Open World Recognition
What this work does
Formalizes recognition with unknown rejection and subsequent category incorporation, deriving bounded open-space-risk conditions and a Nearest Non-Outlier classifier that supports incremental updates.
Connection to discovery
An early operational foundation for detecting unknowns and extending an agent's recognized knowledge over time.
Evidence
Inspected the risk definitions, NNO construction and ImageNet protocol that varies known, unknown and incrementally added categories.
Scope and limitations
Category recognition under supplied features and labels is a bounded precursor; it does not by itself test experiment selection or explanatory mechanism discovery.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.