Hierarchical Multi-Label Text Classification in a Low-Resource Setting

In this thesis we address a multi-label hierarchical text classification problem in a low-resource setting and explore different approaches to identify the best one for our case. The goal is to train a model that classifies English school exercises according to a hierarchical taxonomy with few labeled data. The experiments made in this work employ different machine learning models and text representation techniques: CatBoost with tf-idf features, classifiers based on pre-trained models (mBERT, LASER), and SetFit, a framework for few-shot text classification. SetFit proved to be the most promising approach, achieving better performance when during training only a few labeled examples per class are available. However, this thesis does not consider all the hierarchical taxonomy, but only the first two levels: to address classification with the classes at the third level further experiments should be carried out, exploring methods for zero-shot text classification, data augmentation, and strategies to exploit the hierarchical structure of the taxonomy during training.

Optimisation of chemical lysis protocol for point-of-care leukocyte differentiation

The full blood cell (FBC) count is the most common indicator of diseases. At present hematology analyzers are used for the blood cell characterization, but, recently, there has been interest in using techniques that take advantage of microscale devices and intrinsic properties of cells for increased automation and decreased cost. Microfluidic technologies offer solutions to handling and processing small volumes of blood (2-50 uL taken by finger prick) for point-of-care(PoC) applications. Several PoC blood analyzers are in use and may have applications in the fields of telemedicine, out patient monitoring and medical care in resource limited settings. They have the advantage to be easy to move and much cheaper than traditional analyzers, which require bulky instruments and consume large amount of reagents. The development of miniaturized point-of-care diagnostic tests may be enabled by chip-based technologies for cell separation and sorting. Many current diagnostic tests depend on fractionated blood components: plasma, red blood cells (RBCs), white blood cells (WBCs), and platelets. Specifically, white blood cell differentiation and counting provide valuable information for diagnostic purposes. For example, a low number of WBCs, called leukopenia, may be an indicator of bone marrow deficiency or failure, collagen- vascular diseases, disease of the liver or spleen. The leukocytosis, a high number of WBCs, may be due to anemia, infectious diseases, leukemia or tissue damage. In the laboratory of hybrid biodevices, at the University of Southampton,it was developed a functioning micro impedance cytometer technology for WBC differentiation and counting. It is capable to classify cells and particles on the base of their dielectric properties, in addition to their size, without the need of labeling, in a flow format similar to that of a traditional flow cytometer. It was demonstrated that the micro impedance cytometer system can detect and differentiate monocytes, neutrophils and lymphocytes, which are the three major human leukocyte populations. The simplicity and portability of the microfluidic impedance chip offer a range of potential applications in cell analysis including point-of-care diagnostic systems. The microfluidic device has been integrated into a sample preparation cartridge that semi-automatically performs erythrocyte lysis before leukocyte analysis. Generally erythrocytes are manually lysed according to a specific chemical lysis protocol, but this process has been automated in the cartridge. In this research work the chemical lysis protocol, defined in the patent US 5155044 A, was optimized in order to improve white blood cell differentiation and count performed by the integrated cartridge.

Simulation of URLLC slicing with resource reconfiguration in 5G networks

This thesis seeks to analyse the performance of dynamic slice provisioning in a 5G metro network with the low latency and reliability guaranties. This elaborate highlight the comparison in terms of performance of two versions of a simulator developed in Python based on different models: the Exhaustive research model and Shortest Path First Fit (SPFF) model. It further presents the differences between the dedicated path protection and the shared path protection. This analysis is made through several simulations at different network conditions by varying networks resources and observing the network performances while comparing the 2 models mentioned above. A reconfiguration procedure was implemented on backup resources in the shortest path first fit in order to improve its performance with respect to the exhaustive research which is more optimised. Subsequently, several triggering events was implemented, for the reconfiguration. And a comparison is made between these different triggering events in terms blocking probability, bandwidth at link, capacity at each node, primary and backup bandwidth per slice and backup capacity per slice.