This project is particularly suitable for those students who wish to pursue further study in a PhD in novel, lightweight and efficient machine learning.

The proliferation of the Internet of Things (IoT) has led to the constant generation of massive data from electronic devices. The techniques that make sense of data largely rely on machine learning. However, conventional machine learning, particularly deep learning, demands great computational power and consumer high energy so they are suitably implemented in the cloud and data centres. Many applications require the performance of cognitive tasks at the edge due to security and privacy constraints, response latency and communication cost. The limited energy capacity and computing resources of edge devices make the computationally demanding machine learning algorithms impractical for mass deployment. Therefore, it is crucial to develop computational and energy-efficient algorithms that enable data handling and intelligence embedding on edge devices.

This project will investigate an emerging computing framework Vector Symbolic Architectures (VSA, aka hyperdimensional computing) for lightweight machine learning algorithms for efficiently performing cognitive tasks at the network edge. VSA is a form of brain-inspired computing for representing and manipulating data in a high-dimensional vector space. Unlike classical computing dealing with bits through logical operation and four arithmetic operations of addition, subtraction, multiplication and division, VSA deals with hypervectors through three operations of multiplication, addition and permutation. Its distributed representation and manipulation of information inherently makes the computing robust, scalable, and energy efficient and requires less time and data for training and inference.

In particular, this project will evaluate the latest VSA-based machine learning method with the potential to identify any improvement and extension. An example code for the project is available at https://github.com/eaoltu/hyperseed

Indicative References

Pentti Kanerva (2009) "Hyperdimensional Computing: An Introduction to Computing in Distributed Representation with High-Dimensional Random Vectors," Cognitive Computation, 1(2), 139–159. https://doi.org/10.1007/s12559-009-9009-8

Junyao Wang, Haocheng Xu, Yonatan Gizachew Achamyeleh, Sitao Huang, Mohammad Abdullah Al Faruque (2024) "HyperDetect: A Real-Time Hyperdimensional Solution for Intrusion Detection in IoT Networks," IEEE Internet of Things Journal, vol. 11, no. 8, pp. 14844 - 14856. https://doi.org/10.1109/JIOT.2023.3345279

Evgeny Osipov et al. (2024) "Hyperseed: Unsupervised Learning With Vector Symbolic Architectures", IEEE Transactions on Neural Networks and Learning Systems, vol. 35, no. 5, pp.6583-6597. https://doi.org/10.1109/TNNLS.2022.3211274