Optical content-addressable memories for tree-based machine learning

Le résumé fourni par la source

In this work, we describe the hardware implementation of a decision tree (DT) model using our microring-based timedivision multiplexing (TDM) optical content-addressable memory (O-CAM) architecture, which uses a customized encoding scheme that allows ternary search. Mainstream machine learning models, such as neural networks and transformers, are capable of impressive accuracy when dealing with unstructured data such as images or text, but DTs still outperform them for regression and classification tasks over tabular data. We developed a novel DT mapping scheme for O-CAM and a new inequality encoding technique, enabling the parallel evaluation of inequalities for all nodes of the DT.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.