Adir Rahamim

Title: Adir Rahamim: Inventor in Text Augmentation

Introduction

Adir Rahamim is an inventor based in Netanya, Israel. His patent work includes a computer-implemented approach to text augmentation through dataset reconstruction.

Latest Patents

Rahamim is associated with 1 patent, titled “Text Augmentation Using Dataset Reconstruction.” The method begins with a labeled source dataset containing textual instances across two or more classes. A machine-learning classifier is trained on the dataset and then used to produce hidden representations for a subset of its instances. A trained multilayer perceptron generates soft prompts from those representations, which are supplied to a trained language model to reconstruct the selected text instances.

Career Highlights

Rahamim works at International Business Machines Corporation (IBM). His patent describes a technique combining classification, learned representations, soft prompts, and language-model reconstruction for text augmentation.

Collaborations

The patent’s approach brings together machine-learning classification, multilayer perceptron networks, and language models. The supplied information does not identify additional collaborators.

Conclusion

Adir Rahamim’s patent focuses on using dataset reconstruction and soft prompts to augment text, reflecting IBM’s work in machine-learning methods.