In this paper, we present GrokNet, a deployed image recognition system for commerce applications. GrokNet leverages a multi-task learning approach to train a single computer vision trunk. We achieve a 2.1x improvement in exact product match accuracy when compared to the previous state-of-the-art Facebook product recognition system. We achieve this by training on 7 datasets across several commerce verticals, using 80 categorical loss functions and 3 embedding losses. We share our experience of combining diverse sources with wide-ranging label semantics and image statistics, including learning from human annotations, user-generated tags, and noisy search engine interaction data. GrokNet has demonstrated gains in production applications and operates at Facebook scale.

This paper describes the Incomplete Relational Knowledge Representation, and its support by associative processing. Top-down design of an architecture to support the model (the Generic Relational Model), has yielded functional primitives concerned with the associative derivation of sets from type lattices (Multiple Inheritance Networks). These primitives have a general utility beyond the Knowledge Base area, which is our immediate interest. A subsystem architecture known as the Generic Associative Array Processor has been developed to support the required operations. This has been implemented in VLSI technology to provide parallel processing of associative links via a dynamically configurable connection matrix. The subsystem also provides array processing of the Boolean operations on the resultant sets.