){kind=icon}
){kind=icon}
)
Machine learning within Data Science and AI
Discover our diverse array of Machine Learning projects , showcasing cutting-edge applications and solutions designed to revolutionise various fields.
We also have a selection of projects in other themes with relevance to Machine Learning
Machine Learning Techniques for Exchange Rates Forecasting
Participants
Mr. Ogonna Kafor
Summary
The foreign exchange (forex) market recognised as one of the world’s most significant financial markets is complex, dynamic, and characterised by high volatility, uncertainty and irregularity. Accurately predicting forex rate behaviour is a major challenge for all stakeholders (e.g., traders, investment firms, banks, etc.). This project explores and develops novel machine learning models that offer more accurate and potentially more reliable predictions by addressing weaknesses in current models and adapting to variation in data, changing market conditions and unexpected events.
)
Partners
Publications
[1] T.Alade and O.Kafor, "A Novel FIG-LSTM Ensemble Machine Learning Technique for Currency Exchange Rate Forecasting", to appear in 37th IEEE Canadian Conference on Electrical and Computer Engineering (IEEE CCECE 2024), August 2024.
Markerless Human Motion Estimation and Evaluation
Participants
Ms. Leila Malekian
Summary
Single and multiple video cameras, referred to as markerless motion capture methods, offer a modern alternative to traditional optical marker-based systems. These methods are more cost-effective, versatile, and compatible with advanced algorithms. Additionally, they do not require specific clothing or markers, thereby increasing the potential volume of data collection. Markerless human motion capture is particularly valuable in injury prevention, rehabilitation and other sports and clinical applications. However, despite their many benefits, the accuracy of markerless methods, especially with a single video camera, remains significantly lower than that of gold-standard optical motion capture systems. Given the stringent accuracy requirements of many applications, developing algorithms that enhance precision is essential. Equally important is the selection of an accurate model representation of the human body and the correct measurement of accuracy. Motion signals extracted from any input modality can be analyzed at different levels of abstraction, such as activities, actions, or gestures. This motion data can then be evaluated by assigning a score or quality metric to it. For this project, motion evaluation is conducted for sport activity scoring purposes, specifically focusing on martial arts movements.
The accuracy of human motion estimation and evaluation is enhanced through a range of classic machine learning and deep learning techniques. In classic machine learning, various basic and complex handcrafted features are combined with different models to achieve this. Deep learning methods are employed for both human motion evaluation and pose recovery in cases of occlusion. Additionally, a new training dataset has been developed, consisting of video-based human motion capture using synchronized optical motion capture and monocular video.
Smart environments research facility
Participants
Summary
This project provides funding support for the equipment in 4G/5G and AIoT testbed. We brought GPU servers, 4G small cells, SDN routers, BLE, Zigbee, LoRa, WiFi commutations modules, various sensors, etc. We got a licence band certificate from Ofcom and built a 4G/5G private network testbed on UEA campus. Based on the testbed, we have cooperated with Venari Security Ltd. on 5G encrypted traffic analysis [1]. Other work related to virtual core network can be found at [2], [3].
)
Funding
£100K
Publications
[1] R. Xiong, K.-L. Tong, Y. Ren, W. Ren, and G. Parr, “From 5G to 6G: It is time to sniff the communications between a base station core networks,” in Proc. ACM MobiCom ’23, Demo paper, Madrid, Spain, 2023, pp. 1–2.
[2] C.-Y. Hsieh, Y. Ren, and J.-C. Chen, “Edge-cloud offloading: Knapsack potential game in 5G multi-access edge computing,” IEEE Transactions on Wireless Communications (TWC), voly. 22, no. 11, p. 7158 - 7171, Nov. 2023
[3] Y. Ren, T. Phung-Duc, J.-C. Chen, and F. Y. Li, “Enabling dynamic autoscaling for NFV in a non-standalone virtual EPC: Design and Analysis" JIEEE Transactions on Vehicular Technology (TVT), vol. 72, no. 6, p. 7743 - 7756, Jun. 2023
An intelligent data processing platform for smart manufacturing- an AIoT platform
Participants
Summary
ML and AI have played a significant role in digital transformation. This project aims to develop an AIoT platform, namely AIoTtalk, which is a major enhancement of IoTtalk developed by our Taiwanese project partner. AIoTtalk is a framework of highly integrated data decentralized AIoT platform. In order to achieve this aim, we will first improve the architecture models of original IoTtalk to map into AI-enabled IoT applications. Secondly, we will develop cases study exemplars on essential domains, such as manufacturing disseminate the AIoT benefits. After that, we will build an ecosystem to manage the AIoT model for efficient re-deployment [1].
)
Partners
Funding
Publications
[1] Y.-C. Liang, K.-R. Wu, K.-L. Tong, Y. Ren, and Y.-C. Tseng, “An exchange-based AIoT platform for fast AI application development,” in Proc. ACM Q2SWinet ’23, Montreal, Canada, 2023, pp. 1–10.