RECO-NLP: Low-Power Reconfigurable Computing Architecture for Accelerating NLP.
Abstract
National Language Processing is an application of artificial intelligence that offers applications to companies that need to analyze their data reliably. It stands for Neuro-Linguistic Programming and is based on the idea that how we think, feel and behave is linked to how we use language and our neurological processes. This quality efficiently enables human-computer interaction and allows for the analysis and formatting of large volumes of previously unused data. As per NLP market forecast, post COVID-19, the NLP size was valued at $15,554.31 million in 2021 and is projected to reach $341,530.29 million by 2030, growing at a Compound Annual Growth Rate (CAGR) of 40.9% from 2020 to 2030. This proposal studies different computing strategies to accelerate Bidirectional Encoder Representations from Transformers (BERT), which is an open-source machine learning framework. The proposal will focus on delivering a unique hardware/software accelerator that can serve under physical and battery lifetime constraints. The proposal focuses on building a newgeneration accelerator using the combination of reconfigurable computing (using FPGA platforms) and edge computing. The reason for using FPGA is to benefit from the software/hardware portioning and overlay architecture, which some platforms, such as FPGA, offer without adding any overhead on the battery lifetime. From the academic excellence perspective, this research project will engage students and faculty at a minority-serving institute such as Cal Poly Pomona, which will help the students to learn new top-notch knowledge and technology that will help them in their future careers after graduation. Cal Poly Pomona is part of the 23-campus California State University (CSU) System. Cal Poly Pomona is a minority-serving institute (MSI) based on the enrollment criteria. The BERT NLP accelerator that will be built based on the research outcomes of this research proposal will be tested on unclassified datasets that the NSWC Corona collaborator will provide
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 12, 2025
- Source ID
- N001742310002
Entities
People
- Mohammad Aly
Organizations
- California State Polytechnic University, Pomona
- United States Navy