Backpropagation-Free Continual Learning in Boolean Space for Extreme Edge Intelligence

Read full text Summarise with VIRAT
Authors: Vladimer Khasia
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

The deployment of adaptive machine learning on extreme edge hardware is fundamentally bottlenecked by the computational and spatial complexities of gradient-based optimization and floating-point arithmetic. To circumvent these limitations, we present a backpropagation-free continual learning framework natively engineered for high-dimensional Boolean space ($\mathcal{H}$-space). By projecting continuous, low-dimensional data into a pseudo-orthogonal $8192$-bit geometric space, the proposed methodology replaces energy-intensive matrix multiplications with hardware-native bitwise operations (exclusive-OR and population counts) acting strictly over bit-packed memory registers. Learning is achieved geometrically via bipolar associative memory and dynamic superposition, which inherently bypasses iterative error-correction for novel data. Empirical evaluation demonstrates that the framework achieves competitive predictive performance on high-dimensional sensor streams while exhibiting microsecond-scale adaptation. Notably, the system assimilates a completely novel class via 5-shot continual learning in $0.96$ milliseconds, strictly preventing catastrophic forgetting. Bounded by an exact $\mathcal{O}(C \cdot D/W)$ space and time complexity, this entirely integer- and bit-based architecture provides a highly scalable, energy-efficient foundation for direct field-programmable gate array (FPGA) integration, embedded microcontrollers, and real-time streaming classification across domains ranging from IoT sensor networks to biological motif recognition. The code is available at {\url{https://github.com/VladimerKhasia/booleanclass}}

Fields of Study & Subjects

Machine Learning and ELM Neural Networks and Applications Advanced Memory and Neural Computing

Publication details

Journal / source
Zenodo (CERN European Organization for Nuclear Research)
Publication date
22 March 2026
Language
EN