Partial-update dimensionality reduction for accumulating co-occurrence events

• The input of our task is given by a long sequence of co-occurrence events.
• The goal of our task is to learn similarity metrics give input stream.
• The target similarity between two objects is proportional to their co-occurrence rate.
• We propose a dimensionality reduction that approximates target inter-object similarities.
• Experiment results show that our algorithm learns gradually the target similarity.

This paper addresses a novel problem when learning similarities. In our problem, an input is given by a long sequence of co-occurrence events among objects, namely a stream of co-occurrence events. Given a stream of co-occurrence events, we learn unknown latent vectors of objects such that their inner product adaptively approximates the target similarities resulting from accumulating co-occurrence events. Toward this end, we propose a new incremental algorithm for dimensionality reduction. The core of our algorithm is its partial updating style where only a small number of latent vectors are modified for each co-occurrence event, while most other latent vectors remain unchanged. Experiment results using both synthetic and real data sets demonstrate that in contrast to some existing methods, the proposed algorithm can stably and gradually learn target similarities among objects without being trapped by the collapsing problem.