Neurally-Inspired Integration of Communication and Cognitive Computation in Hyperspace

University Of California-Irvine, Irvine CA

Investigators

Abstract

The next generation of communication systems holds the potential to enhance every aspect of our lives. From enabling seamless global connectivity and telemedicine advancements to transforming transportation, fostering education and entrepreneurship, supporting disaster response, and driving economic growth. However, current computer systems face a challenge where communication and learning applications operate independently, leading to inefficiencies and delays caused by the need to coordinate between these separate layers. Merely improving battery life in mobile devices is insufficient to keep up with the growing demand for efficient processing, especially considering the rapid advancements in machine learning. This project seeks to address these fundamental issues by pioneering advanced technology for communication that seamlessly integrates channel coding with machine learning. By bringing these two domains together, we aim to achieve hyper-reliable communication and efficient machine learning processing. Moreover, we aim to develop future learning systems that can operate flawlessly even in noisy and unreliable network environments. This approach has the potential to provide users with the benefits of neurally-inspired learning models across various network, thereby significantly enhancing the efficiency and robustness of these systems. In addition to its technical advancements, this project recognizes the importance of education, diversity, and broader societal benefits. This project aims to establish a new exploratory undergraduate research program, fostering early research scholars who will contribute to the field of intelligent networks and computing systems. Furthermore, international collaboration will be a key component, allowing us to engage with global experts and share knowledge to advance the field as a whole. This research project presents a neural-inspired network system for robust and efficient data communication and information processing. It achieves this by redesigning communication and machine learning using a unified neural mathematical foundation with high compatibility. The main contributions of our proposal are as follows: (1) Designing rigorous communication schemes that work with the high-dimensional representation of transferred data in a mathematically tractable manner. This communication scheme consists of encoding/decoding methods that leverage redundant and holographic neural representations to achieve ultra-efficient and robust data communication. It also provides concrete network-level simulation and optimization to ensure the desired latency and responsiveness. (2) Fundamentally merging channel coding and learning by directly computing over transmitted data, eliminating the need for costly iterative data decoding. This eliminates the inter-layer overhead between communication and computation procedures that exists in prior communication schemes. (3) Designing a hardware platform that offers high parallelism and efficiency for both communication and computation. The proposed highly optimized hardware will automatically map applications to FPGA, equipped with a transceiver module, and leverage the parallelism provided by these platforms. Our framework will be evaluated using both complete simulation environments and real-world deployment in communication and application infrastructures. The prototype will be fully released under an established open-source library for public dissemination. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.