Abstract
ZebraLSTM is a novel Long Short-Term Memory (LSTM) architecture designed to optimize the performance of sequential data processing tasks. By dividing the hidden state into multiple parallel stripes, ZebraLSTM enhances computational efficiency and scalability, particularly when leveraging modern GPU architectures. This document outlines the design, implementation, and potential applications of ZebraLSTM in various domains, including natural language processing, time-series analysis, and drug design.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Gagnon, Renee, "ZebraLSTM: An Advanced LSTM Architecture for Efficient Sequential Data Processing", Technical Disclosure Commons, (July 01, 2024)
https://www.tdcommons.org/dpubs_series/7148