COHORT.3

2021 – 2025

Modal Title

Alec Parise

Student

Project Title

Human-In-The-Loop Rule Based Feedback Towards Interactive Deep Learning

Project Description

As technology evolves, the more common it has become to encounter cases of faulty systems and data that have a deep impact on people’s lives. The most prominent ones are in the field of facial recognition and state surveillance in which innocent people are being arrested, gender bias in work recruitment and credit in which women are eliminated from recruitment processes and are denied credit because of “sexist” data sets, people dying in hospital triage because of the use of wrong data sets for classification models, and the list goes on. First and foremost: Do people know what Artificial Intelligence (AI) is? Do they know when they’re engaging with it? The answer is hardly certain, therefore how can they be fully aware if they are being harmed by AI? What is currently seen in practice are algorithmic audits which are key when tackling the source of the problem, but there is still a long way to go until these procedures are fully established and largely implemented. It’s still depending on companies and governments to search for qualified people to conduct the audits, but what happens when there’s no interest? As for what this project proposes, is to listen to individual experiences, one-by-one, trying to build a network and search for patterns. We will take an ethnographic approach to data sets, as we do with people’s narratives, and that will be our starting point. This project initially consists of figuring out the impact AI/ML systems have/ are having/had on people’s lives and to further understand precisely how they are/were affected. For that, interviews will be conducted and if possible recorded, so that we can produce not only oral but also visual records of people telling their stories. In the next phase of the project, we will run topic modelling and Latent Dirichlet Allocation (LDA) which seems to be a powerful tool for recognizing such patterns in discourse and also innovative in ethnographic research. When we draw from people’s experiences we are able to build a more relatable narrative which can help us build systems that take them into consideration. There is a necessity of building bridges between computer scientists, data scientists and social scientists in order to create AI systems, machine learning models and data sets that can be used in a non-harmful way.

Modal Title

Ana Paula Moritz

Student

Project Title

“Robots or something?” : An ethnography of Artificial Intelligence and its effects on society

Project Description

As technology evolves, the more common it has become to encounter cases of faulty systems and data that have a deep impact on people’s lives. The most prominent ones are in the field of facial recognition and state surveillance in which innocent people are being arrested, gender bias in work recruitment and credit in which women are eliminated from recruitment processes and are denied credit because of “sexist” data sets, people dying in hospital triage because of the use of wrong data sets for classification models, and the list goes on. First and foremost: Do people know what Artificial Intelligence (AI) is? Do they know when they’re engaging with it? The answer is hardly certain, therefore how can they be fully aware if they are being harmed by AI? What is currently seen in practice are algorithmic audits which are key when tackling the source of the problem, but there is still a long way to go until these procedures are fully established and largely implemented. It’s still depending on companies and governments to search for qualified people to conduct the audits, but what happens when there’s no interest? As for what this project proposes, is to listen to individual experiences, one-by-one, trying to build a network and search for patterns. We will take an ethnographic approach to data sets, as we do with people’s narratives, and that will be our starting point. This project initially consists of figuring out the impact AI/ML systems have/ are having/had on people’s lives and to further understand precisely how they are/were affected. For that, interviews will be conducted and if possible recorded, so that we can produce not only oral but also visual records of people telling their stories. In the next phase of the project, we will run topic modelling and Latent Dirichlet Allocation (LDA) which seems to be a powerful tool for recognizing such patterns in discourse and also innovative in ethnographic research. When we draw from people’s experiences we are able to build a more relatable narrative which can help us build systems that take them into consideration. There is a necessity of building bridges between computer scientists, data scientists and social scientists in order to create AI systems, machine learning models and data sets that can be used in a non-harmful way.

Modal Title

Badrinath Singhal

Student

Project Title

Performance and Scalability of Recommendation Algorithms

Project Description

Recent developments of new algorithms for recommender systems have followed two quite distinct tracks: (1) highly scalable linear algorithms, such as EASE and SLIM, that have been shown to work well on very large and sparse datasets. (2) Deep models, what learn user and item embeddings through a neural network architecture, that is computationally intensive to train. Another approach which is much less well explored in the state-of-the-art is the Bayesian approach, in which a generative model of the recommendation process is posited and an inference algorithm such as Markov Chain Monte Carlo or Variational Inference is applied to learn full distributions of the parameters of the model. Similar to deep models, a major challenge for Bayesian approaches is the computational intensity of the training process. This project will explore novel algorithms for recommendation, focusing on performance and scalability. We will consider whether a Bayesian model with tractable inference is feasible for the recommendation setting and examine the integration of a Bayesian approach with deep models or simple scalable models. As well as tackling performance, other qualities of the recommendation will be considered, such as diversity and novelty.

Modal Title

Bulat Maksudov

Student

Project Title

Explainable AI for Medical Images

Project Description

Despite the success of Deep Learning in natural language processing and computer vision, the lack of human interpretability hinders their use in high-stake decision making. The aim of the proposed research idea is to tackle the problem of generating deep explanations that rely on multimodal data, and where the interpretation of some specific features in one modality can support explanation of features in another modality. The idea is to explore the use of techniques for semantic linking of causal and relational structures extracted from deep representations to identify how they correlate multimodal representations. One possible way of doing that would be to leverage functional graphs representing the neural activity within the deep network, and use probabilistic graphical models, statistical relational learning and/or link prediction to predict and validate semantic connections across modalities. The ultimate goal should be to not only explain the outcome of deep learning models, but also to link concepts from one modality to another to generate better explanations. One application scenario that can be investigated to demonstrate the approach is multimodal diagnostics (eg. triage).This project aims to explore several research challenges regarding the usage of AI for medical imaging and the challenge of transparency and explainability: how can our model provide actionable and clinically significant output? What are the differences between the decision-making process of radiologists and medical imaging models? How can we incorporate additional data and knowledge to affect trust and interpretability of the output of the model?

Modal Title

Chris Vorster

Student

Project Title

Computationally efficient Machine Learning for Compressed Video in Edge Devices

Project Description

A clear trend in Internet of Things (IoT) research is to move intelligent decision-making away from centralised cloud servers and towards low-power, low-latency smart devices close to the data source and user, while supported by centralised oversight and configuration control. Challenges in moving Machine Learning (ML) from cloud to edge devices include the relatively constrained computing power and memory of edge devices, which in turn constrains the accuracy of ML models.In such limited settings, Video Object Detection is frequently accomplished by performing Image Object Detection on a per frame basis. Real-time video streams therefore need to be decoded into individual images, which is inefficient. This approach also neglects temporal information.The first avenue of this research is to investigate the feasibility and accuracy of Object Detection by applying deep learning models directly to key frames and using the non-key frames as a proxy for motion. Objects can be detected and recognised using the spatial information in the key frames, aided by the object cohesiveness which motion provides through the non-key frames. This research will build on the work done by Wang et al. in Fast Object Detection in Compressed Video (2019) and Real-time and accurate object detection in compressed video by long short-term feature aggregation (2021).Further research directions will include the creation of a dataset that combines compressed video and neuromorphic camera data along with the application of ML on such data. Neuromorphic cameras are event-based, low-power sensors that detect changes in brightness per pixel. The similarity between neuromorphic camera’s representation of reality and that of compressed video will be investigated.Possible research avenues include memory utilization, novel model architectures and training data transformations. Additionally, this research will develop new architecture designs, techniques and models for implementing ML-based applications on edge devices in order to efficiently perform ML on IoT devices.Use cases of this work will focus on computer vision applications, being data intensive processes requiring ML support, which is particularly challenging for edge devices. Output will be evaluated against current state-of-art deep learning platforms, including Google Cloud ML Engine.Applications of this work are wide-ranging, including automotive driver assist, predictive maintenance, intruder detection and wildlife monitoring. Any application with scene understanding from cameras could benefit from this work.

Modal Title

Di Meng

Student

Project Title

The flavour of disorder: predicting intrinsically disordered regions in proteins by Deep Learning

Project Description

Proteins are the basis of life and over the last few decades we have learned much information about them through genome sequencing projects and other massive-scale experiments. However important aspects of proteins such as their structure and function remain elusive and the experimental techniques devised to reveal them have not scaled up as quickly as the techniques that elucidate their sequence or expression. Nowadays we know the sequence of well over one hundred million proteins, while the structure/function is known for less than 0.1% of these. For decades the paradigm that proteins formed rigid, stable structures was essentially unquestioned, while it is now clear that many proteins only partly fold to a native regular structure or are normally completely unfolded or varied between folded and unfolded (semi-unfolded). By some estimates, up to 20% of amino acids in known proteins are in a disordered state. We currently have datasets comprising over 180,000 proteins for which disorder information is known in some form. The aim of this project is the prediction of disordered regions in proteins. The problem will be tackled by an array of Deep Learning techniques, which can learn the likely locations of disorder or semi-disorder from examples of proteins in which these locations are known experimentally. Also, we could dig into these locations to investigate disordered binding and semi-disorder variation. Upon success, the results of the project may feed into the online Distill servers and improve the quality of their results. The Distill servers are a widely used tool, with millions of queries served originating from over 100 national and transnational internet domains from all over the world, and even a marginal improvement of their performances would benefit a large pool of scientists world-wide and help them further their research on biology, biotechnologies, and drug design.

Modal Title

Dzelila Siljak

Student

Project Title

A Unified Framework For Automated, Accurate and Flexible Knowledge Graphs from Free Text

Project Description

Research in Machine Learning AI has made great advances but recently, there has been much discussion centering around the issues with state-of-the-art models: lack of interpretability and transparency, inability to generalize and reason in unseen situations, and, frequently, the need to train these models on large labeled datasets that are often difficult and expensive to generate at the scale required. There is increasing interest in integrating symbolic knowledge representation and reasoning methods into Machine Learning solutions in order to tackle these issues and create adaptable systems that can be applied to a variety of domains and settings. Knowledge Graphs have been achieving increasing visibility in the research community as a form of structured representation of information. Integration of Knowledge Graphs into downstream tasks has already shown great potential in use-cases such as Question Answering and recommendation. A major bottleneck, however, is the still-unsolved challenge of automated creation and curation of Knowledge Graphs that are accurate, can be maintained with minimal manual intervention, and balance the tradeoff between adherence to design requirements and the flexibility necessary for integration of new knowledge and generalization. Our work focuses on addressing the Knowledge Graph creation and curation bottleneck, with specific focus on extracting knowledge from free text. We intend to tackle this by considering all aspects of end-to-end Knowledge Graph construction and application to downstream tasks: domain discovery, schema design, information extraction for KG population, Knowledge Graph completion, evaluation, as well as maintenance and downstream tasks. These aspects are usually considered in isolation in research work, whereas we propose to approach the bottleneck problem via a more unified framework that pushes the boundaries of existing methods individually and establishes end-to-end systems that benefit from mutual interaction and feedback. The first contribution addresses the bottleneck of Knowledge Graph construction and curation with particular attention to knowledge extraction and Knowledge Graph population from free text. Our approach consists of incorporating existing linguistic and domain-specific knowledge bases for downstream linguistic tasks, as well as enriching distributed semantic representations with syntactic information through recursive structures and neuro-symbolic reasoning. The goal of our efforts in this part of the project is to establish a two-way relationship between NLP methods and symbolic knowledge representation and reasoning.
Modal Title

Eliane Birba

Student

Project Title

Machine Learning for Better Outcomes in Hematopathology Using Platelet-based Biomarkers and Mass Spectrometry Data

Project Description

The application of Artificial intelligence (AI) in healthcare has increased significantly in recent years, driven mainly by an abundance of data and powerful, accessible tools. Millions of blood cells are evaluated for hematological diagnostics by clinicians every day and this area offers significant opportunities for AI and machine learning. AI offers many promising tools to clinicians working in hematology to speed up and automate blood analysis. The proposed research seeks to employ AI in the analysis of hematopathology. The first objective is to develop a new algorithm for detection of Preeclamptic toxemia (PET) in pregnant women. Annually, PET claims the lives of 50,000 mothers and 500,000 babies and accurate, easily deployed detection tools do not exist. Therefore, there is an urgent, unmet challenge to develop accurate risk stratification tools for PET. This work will develop a new solution for risk stratification in PET. To achieve this, we will use a machine-learning algorithm to build a reliable and accurate test for preeclampsia, with a simple and easily interpretable score that can be deployed and implemented into widespread clinical use. This project will be a collaboration of the larger SFI-funded AI_PREMie project and take advantage of data collected and collaborator expertise within that project. The second objective is to apply machine learning to mass spectrometry (MS)-based proteomics analysis. This is an emerging area for application of machine learning techniques and there are a lot of research opportunities. Proteomics’ study is crucial for drug development, early diagnosis, and monitoring of diseases. One main challenge is that the proteome or the set of proteins in a cell/tissue/organism varies from time to time. Therefore, AI and machine learning methods can help fast and accurate protein pattern recognition and classification. This strand of the work will develop new techniques for the application of ML techniques to proteomics data, with a particular focus on making them accessible to clinicians

Modal Title

Faithful Onwuegbuche

Student

Project Title

A Deception Technique for Adaptive Intrusion Detection

Project Description

The rapid rise of the digital economy and the internet is driving growth of businesses, but it is also introducing new cyber security risks. The Accenture 2020 state of cybersecurity report reveals that the three areas of cybersecurity protection with the largest increases in cost are network security, threat detection and security monitoring. To help mitigate network based inside and outside attacks, researchers have developed honeypots which are deception defenses used to divert the attention of attackers from the real systems or networks and to analyze attacks methods and patterns of activities. They are also used to educate security professionals and support network forensics. However, traditional static honeypots can be detected easily using anti-honeypot toolkits, such as honeypot hunter, since they utilize a fixed configuration and response. When a honeypot is detected, an attacker can tamper with the evidence collected by the honeypot and attempt an attack on the real system. To help overcome these weaknesses, researchers proposed dynamic honeypots, which can change their configuration and can make it more difficult for an attacker to detect where valuable assets are located. Dynamic honeypots are usually deployed in a centralized host to support automation. However, this host, if compromised, can lead to the breakdown of the real system. Therefore, blockchain can be used to address this problem since it features distribution and decentralization. Due to the decentralization property, every network node disperses the computation load and has better robustness. The advantage of blockchain relies on the fact that data cannot be tampered with since any change would be revealed by the nodes which are connected to the network. Also, if one host is compromised the same information is still held by other hosts in the network. Therefore, honeypots and honeynet deployed with blockchain integration can better support network forensics as they can prevent fraud and data theft with more auditable features. The objective of this project is to develop a novel deception technique for adaptive intrusion detection. The proposed system will implement honeytokens that redirect the attackers from the real server or network to the blockchain based honeynet, in order to trap and then track the attackers’ activities, patterns and methods. This information will be used to update the intrusion detection system knowledge through online machine learning. Additionally, the honeynet will store data that could potentially serve as digital evidence during forensic investigations or provide information about a security incident.

Modal Title

Gargi Gupta

Student

Project Title

Post hoc explanation for RNNs using state transition representations

Project Description

AI and advanced machine learning techniques have had a significant impact on several facets of our life in recent years, taking over human positions in a variety of complex tasks. In domains as diverse as healthcare, banking, justice, and defence, their applications have had great success. Deep neural networks (DNNs), such as convolutional neural networks (CNNs) and recurrent neural networks (RNNs), have gotten a lot of attention in recent years among these machine learning techniques. Since then, they have demonstrated unparalleled performance in tasks such as speech recognition, image recognition, natural language processing, and recommendation systems, outperforming humans in learning. However, it is observed that these best performing models are way too complex, abstract and opaque due to their complex deep architecture and non-linearity. Henceforth, they lack explainability and reliability. As they do not justify their decisions and predictions, it is difficult for humans to trust them. Unsurprisingly until recently, state-of-the-art CNNs, RRNs and other deep learning models, in general have been commonly regarded as “black boxes” or “black-box models”. Building trust in the deep learning model by validating its predictions and ensuring that it works as predicted and dependably on unseen or unfamiliar real-world data is unquestionably vital. In critical domains such as healthcare applications and autonomous vehicles, a single incorrect decision can have a catastrophic effect on society. Understanding, analyzing, visualizing, and explaining the rationale behind the model’s judgments and predictions is critical for ensuring the model’s reliability and understanding the model’s potential limitations and faultsA recurrent neural network (RNN) is a type of artificial neural network that uses sequential or time-series data. Recurrent neural networks, like feedforward and convolutional neural networks (CNNs), use training data to learn. They are distinguished by their “memory,” which allows them to use information from previous inputs to influence the current input and output. One of the challenging tasks with RNNs is to comprehend and evaluate their behavior. This is because it is difficult to understand what exactly they learn and also, they tend to work poorly on sequences requiring long-term memorization, despite having this capacity in principle. Many scholars have previously investigated a variety of strategies to address the aforementioned difficulties in recent years. The fundamental goal of this study is to investigate alternative methods for extracting interpretable state representations such as graphs, finite state machines, deterministic finite state machines from trained recurrent neural networks. The findings of this study will be useful across various domains in the industry and research community in order to provide better explanations to society for the deep learning applications they build and also comply with the GDPR rules.

Modal Title

Jean Francois Itangayenda

Student

Project Title

Bias in Financial Lending Models

Project Description

The research will delve into machine learning models and their perceived and received bias in lending financial transactions. By perceived bias, we want to understand how bias is seen and analyzed by model developers (for example, through their own judgment, model builders, with an intent of increasing fairness in their model, can exclude certain variables, features and categories in order to obtain a desired output). We want to understand how this act of removing certain items from models can “bias” or “influence” their outputs; by received bias, we mean bias that a model might communicate to its own process (training, test) – through data corruption, noise, etc.) We want to understand from a socio-technical perspective how this received bias —whether from model training, datasets or the models themselves—affects model outcomes. The project will be two-sided, looking at both the social impacts of bias in machine learning algorithms that predict and recommend loan refusal/acceptance to clients of financial institutions, the definition(s) of fairness in financial systems with regards to lending activities, the socio-technical forces that may or may not unknowingly impact data collection (for example, causing its corruption) and training; the technical demands of developing and running these models (socio-technical cost-benefit of these models, how are they deployed, how the data is collected, who or what controls these machines that run models (financial institutions, governments, etc.) and how ultimately all these factors impact these models’ outputs.

Modal Title

John O'Meara

Student

Project Title

Development of a Novel Intrusion Detection System and Architecture-specific Datasets in Software-Defined Networking

Project Description

Software Defined Network (SDN) technology allows for more efficient scaling of networks though central programming of network behaviour using software applications with open APIs. The separation of the control plane from the forwarding plane of the physical hardware allows for a consistent network management strategy regardless of network size or complexity. Unfortunately, as with any new technology, these benefits are accompanied by a host of new threats with the revised infrastructure providing new attack vectors for malicious actors intent on penetrating and/or disrupting network activity. The relatively nascent status of SDN technologies makes development of effective Intrusion Detection Systems (IDS) difficult. There is a lack of available SDN specific datasets, resulting in the deployment of IDS software, which has been developed using unsuitable data collected from traditional networks and hence, ignoring the architectural differences of SDN networks. The aim of this research is to focus specifically on the novel architecture of SDN technologies and to develop an appropriate IDS framework that is tailored to the unique architectures of SDN, effectively identifying and blocking attacks that focus on SDN-specific characteristics, in addition to the range of attacks to which standard networks are prone. The intended research will focus on generating new SDN specific datasets by deploying different SDN architectures, both in the virtual format and using physical devices, allowing for the collection of more intrinsic data. Effective IDS can then be developed by training Machine Learning (ML) models on these new datasets. Standard Supervised ML Models as well as unsupervised Deep Neural Network (DNN) and Reinforcement Learning (RL) models will be developed and evaluated. There are a series of expected challenges to be addressed within the proposed body of work. The specific points of difference in architecture between traditional networks and SDN networks will have to be identified and attack vectors designed and/or replicated. Appropriate test bed architectures must be chosen and implemented. The choice of ML model will depend on the dataset structure and will need to be tailored to this specific use case. Finally, appropriate validation systems will need to be crafted to comprehensively test the effectiveness of candidate IDS frameworks once developed.

Modal Title

Joyce Mahon

Student

Project Title

Integrating Machine Learning and Artificial Intelligence into Pre-University Education

Project Description

Artificial intelligence (AI) and Machine Learning (ML) technologies are rapidly generating new possibilities for all industries. For instance, it is now possible to automatically generate subtitles for videos, enabling people with hearing loss to use video chat tools that were previously unavailable to them, with little cost. However these technologies can also be used for more nefarious purposes. For example, automating the generation of malicious online social media content used to undermine democratic processes. The contrast between these two use cases of the same technologies is a perfect illustration of how new ML and AI technologies are having a huge impact on our lives, often in ways that raise a lot of technological, legal, ethical and societal issues. Our current and future pre-University students will grow up in a world where these technologies are commonplace, and they will develop the next generation of these technologies. It is important that they not only understand how they work, but also are equipped to make informed judgements about what technologies they want in their lives and society, and how they would like to use them. This work will expose a wider and more diverse audience to ML and AI tools such as analytics, speech recognition, and natural language processing; by developing materials and strategies for their use across a range of disciplines, building opportunities for students and teachers to see these tools in action and become involved in their design and application. This work also has a secondary aim to improve diversity in computer science education. The research questions that this project will address will include: • What are the current opinions of Irish second level students on what ML and AI should be used? • What are the best ways to engage with Irish second level students on the topic of ML and AI?

Modal Title

Laura Dunne

Student

Project Title

Bus Network Optimisation with Machine Learning

Project Description

Buses are a vital component of an urban environment, and shifting away from private cars towards public transport is essential in minimising our environmental impact and creating sustainable cities. The UN Sustainable Development Goal 11 seeks to “Make cities and human settlements inclusive, safe, resilient and sustainable”. Specifically, target 11.2 states that cities should expand public transport. Unfortunately, there is a trend away from public transportation and towards private cars, due to passenger dissatisfaction with the public transport networks. However, with increased urbanisation, enduring widespread use of private cars is unsustainable, and we must make bus transport an attractive option for passengers. Many factors influence a passenger’s transport choices, but convenient routing options and reliable service are frequently reported unmet needs. Unfortunately, there are physical and financial limits on the service provided, so it is crucial to optimise the resources available to provide the best possible service. The proposed research seeks to provide better scheduling and better route design by applying machine learning (ML) to several under-exploited areas in the bus transit domain. Researchers have demonstrated that ML can improve the efficiency of public transport, and the focus to date has been on the application of various ML algorithms. However, the results are often conflicting, and the experiments are usually conducted on a single bus route in a single city. We propose to examine a whole network of buses and also to attempt to validate the transferability of our experiments on unseen routes, ideally from an unseen bus network. We also plan to address the conceptual model of the bus network and how the network is structured before ML modelling, and how this conceptual model interacts with various ML algorithms. Chokepoints are a significant factor that makes bus transport less reliable. Chokepoints cause bus bunching, which has been shown to impact severely upon the passenger’s service. Analysis has demonstrated that chokepoints in bus networks are caused by physical constraints like signalised intersections or bridges and dynamic factors such as weather or school collection times. We propose to work with OpenStreetMaps data to analyse features that impact bus reliability and train ML algorithms that can predict optimum bus routing. By applying ML to the bus transport domain, we hope to add knowledge that will help optimise bus networks.

Modal Title

Liliya Makhmutova

Student

Project Title

Dialogs: stylistic tuning and personalization

Project Description

Have you ever noticed that human behavior changes depending on surroundings and people nearby? Do you use different language styles with your friends, colleagues, or boss? As AI impacts more and more on multiple spheres of our life, we want it to be more tuned to our preferences and adapt as a human being. We want AI to respond in appropriate style and show us relevant information. We can argue that there are three levels of personalization: 1. Low level. This level of personalization is focused on low-level grammatical structures (choosing between simple or complicated words, sentence construction, etc.). 2. Medium level. Medium level focuses on semantics and content. It is at this level that AI decides how much information to show to a user and how many words he or she would need for this. 3. High level. At this level AI should be focused on what to say and how often to address a user and customized to user’s preferences. Which means totally or partially understanding of human personality. My Ph.D. will focus on optimizing conversational AI and conversational systems at a low level, and I will try to use some level 2 concepts. So, the main challenge I want to solve at these levels is to provide the user with relevant content. On the fly and in the context of dialogue systems. A good approach to tackling this problem is to use some basic pre-trained model, and fine tuning is so that we learn more and more about the user (similar to recommendation systems). Another approach that can be used is described in the article “PROTOTYPE-K-STYLE”: Create dialogs with style editing in RAM Su et al. We can also consider stylistic customization using GAN, but there are many problems due to the fact that the text is discrete in nature (therefore differentiation is not possible), however there are some heuristics that can solve this problem, so it is worth considering. I also want to look at the explainability of conversational AI, since this topic is closely related to human interaction. I believe that this work will make the interaction of users with dialog systems more convenient and natural, as well as help to find more relevant information easily and faster.

Modal Title

Manuela Nayantara Jeyaraj

Student

Project Title

Using ML to Signal Gender and the use of Gendered Language

Project Description

The focus of this project is to use machine learning and natural language processing to develop automatic techniques for identifying gender issues and bias in text content. There are a variety of application areas where such techniques can be useful. In 2018, Amazon scrapped the use of their AI internal recruitment model which showed significant bias against women. The model had been trained on the applications and CVs of successful applicants, most of whom were male, hence, it ‘learned’ that successful candidates were typically male. In the recent Labour leadership election in Britain, an analysis of the language used in news articles about the candidates showed discrepancies related to their gender in how they were described. The single male candidate was more likely to be discussed in terms of professional employment, politics and law and order and the two female candidates were much more likely to be discussed in terms of their families, in particular their fathers. Earlier projects in this area have used techniques pioneered by Google to help identify gender issues in news articles and to detect racist sentiment. The approach is based on the idea that gender attribution relies on language use, not on language itself; therefore, there are many other factors which should also be considered when determining who is being referred to in a text. Therefore, the inclusion of women’s representations in text can be argued to be not only important for simulating real-life occurrences, but also valuable as it allows us to understand how perception and social roles influence language use. Gender stereotype hypotheses about textual content tend to situate language use within a wider discourse about gender differences and the ways that they are constructed. Thus, the goal is that providing recommended linguistic modifications and positive reinforcement to authors about written text will influence and change behaviour. Signaling text content that suggests gendered language or is gender-biased can encourage and influence writing behaviour that is gender neutral. Hence, this project explores methods in supervised machine learning and natural language processing related to gender bias in text and gendered-language identification and prediction. The model will harness stylometric features, gender-specific language patterns, discourses of gender difference and principles of cognitive perception about an author’s identity and use NLP techniques to identify phrases, language, constructs or patterns in writing that signal use of textual content with gender bias or gender stereotyping.

Modal Title

Mark Germaine

Student

Project Title

Early detection of gestational diabetes risk using explainable machine learning models

Project Description

Gestational diabetes mellitus (GDM) is defined as glucose intolerance of varying severity and is present in ~12% of Irish pregnancies (O’Sullivan et al. 2011). Women with GDM are at increased risk of adverse health and delivery outcomes leading to complications, including stillbirth, premature delivery, macrosomia, fetal hyperinsulinemia, and clinical neonatal hypoglycemia (Dunne et al. 2012, Danyliv et al. 2015, Gillespie et al. 2013). Consequently, maternity care costs for pregnant women who develop GDM are 34% higher than the average pregnancy (Gillespie et al. 2013), whereas neonatal costs in the first year of life are typically higher than those in offspring of mothers without GDM (Chen et al. 2009).GDM is most commonly diagnosed during weeks 24-28 of the pregnancy and is diagnosed via an oral glucose tolerance test (OGTT) which is a simple test that measures blood glucose response over a period of 2 hours in response to the ingestion of 75g of glucose (ADA). Effective intervention in GDM has beneficial effects for maternal and neonatal outcomes (Damm et al. 2016, Plows et al. 2018), thus if machine learning models can accurately predict the risk of GDM at an earlier date, this may allow for earlier intervention from clinicians, thus potentially reducing health risks to both mother and child, reducing the risk of complications during pregnancy and saving money in the treatment of GDM.Some studies have tried to assess the risk of GDM by measuring fasting blood glucose during the first trimester, however the results have not been promising (Zhu et al. 2012). Further to this, researchers have been attempting to predict the risk of GDM from much earlier in pregnancy, from 8 to 20 weeks in Chinese women (Zheng et al. 2019) using simple models such Bayesian logistic regression. More recently, researchers have applied simple and advanced machine learning models on variables extracted from medical records in China in an attempt to predict GDM (Wu et al. 2021). Whilst some of these models have very high specificity, sensitivity varies between the models applied and the best sensitivity achieved was 70% using a deep neural network. In Ireland, routine care appointments are scheduled between weeks 8-12, thus the variables for the model will be collected during this period as this is when standard care data is collected and thus more data available. The ideal scenario being that the model would have a strong predictive value by week 12 of the pregnancy, half of the time usually taken to detect GDM.The aim of the current research project would be to explore the applicability of machine learning models for predicting the risk of GDM in pregnant women in Ireland.The objectives of this research are to:1) explore the utility and applicability of machine learning models on predicting the risk of GDM based on variables collected during routine pregnancy care.2) assess the efficacy of the machine learning in practice if the results show high accuracy.3) try to achieve this using explainable machine learning
Modal Title

Michael O'Mahony

Student

Project Title

Multi-Modal Generative Models of Stylistic Tuning — Making Personality more Personal.

Project Description

The world is progressively globalising. As a result, many people want to learn new languages to exploit travel, business, and cultural enrichment opportunities. Unfortunately, not everyone has easy access to language learning classes, and they can be costly for people who do. The amount of virtual content and services people are using is increasing year to year. The rising volume of virtual services and content paired with the finite amount of spare time that all users have makes the selection of services and content within services more critical. Many services can be provided or supported by automatic text generation. There are support chatbots, virtual assistants, data to text generators, joke generators, and language tutors. Many of these advancements in text generation had become more feasible once deep neural networks were popularised and applied to the area. Personalisation applied to these services can make them more relevant to individual users. One fruitful application domain of these areas is language learning. An intelligent virtual tutor can service people regardless of location a lot cheaper than a human tutor. On the other hand, some people may not relate to a text generator like a human tutor. However, an intelligent virtual tutor that can adapt to user personality in a continuous fashion can become more relatable to the user. Another approach that may promote relatability is to combine a virtual reality avatar that can perform simple gestures in line with the generated text. The personality of a virtual tutor can be considered a form of style. This style can be customised to suit a user better. Generative adversarial networks or GANs are a neural mechanism often used to tune the outputs of media. Text can be passed through a network and tuned to provide a custom output. A research challenge would be to make these outputs fluid, dynamic, and automatic. We could combine GANs with other generative models to simulate personality for the language tutor. User studies would likely be used to evaluate this research. This project links to the ML-Labs challenge area of machine learning in language, a fundamental area of machine learning.

Modal Title

Nam Trinh

Student

Project Title

Machine Learning approaches to characterization of human decision making under uncertainty

Project Description

Modal Title

Ramin Ranjbarzadeh Kondrood

Student

Project Title

An attention-based mechanism for brain tumor segmentation using four modalities

Project Description

Brain tumor localization and extraction from magnetic resonance imaging (MRI) is a vital task in a wide variety of applications in the medical field. Current strategies demonstrate good performance on Non-Contrast-Enhanced T1-Weighted MRI, but this is not true when confronted with other modalities. Each modality represents different and vital information about the tissue we are working on. So, in this proposal, we propose an algorithm based on four modalities T1, T1c, T2, and FLAIR for segmenting the tumor region with a high rate of accuracy. To increase the efficiency of the model and decrease the evaluation time, a powerful pre-processing approach for removing the insignificant areas of the brain is used. Also, to improve the segmentation result of discrimination between internal areas of the tumor, an attention-based mechanism is used. We will use the BRATS 2018 dataset which comprises the Multi-Modal MRI images. Each patient’s sample in the dataset has the dimensions of 240×240×150 and were annotated by specialists.

Modal Title

Robert Foskin

Student

Project Title

A Reinforcement Learning Approach to Continuous Measurement-Based Quantum Feedback Control

Project Description

Reinforcement learning has had proven success in the domain of classical control and there has been a recent surge of work investigating its application in quantum control problems however not much work has focused on utilizing continuous measurement when training the agent. This is because the act of measurement on quantum systems is fundamentally different to that of the classical counterparts and introduces a number of unique challenges. The goal of this project will be to investigate the application of reinforcement learning to the control of quantum dynamics using continuous measurement-based feedback techniques. A key objective of this project will be the development of novel techniques for quantum state representation in reinforcement learning algorithms and further research in this area has the potential to impact the development of near term quantum technology, including a fault-tolerant quantum processor. In the area of feedback-based control, current methods have not shown sufficient progress. This is because such applications to quantum systems quickly become intractable for standard optimal control techniques when quantum feedback leads to an exponential increase in the search space. Analytical approaches are also difficult to realise consistently for quantum systems in experimental settings. Due to the presence of noise and de-coherence, optimal dynamics in an experimental system diverges from that of the model used when optimizing the control strategy. This is especially true in quantum feedback control where the act of observing the system continuously introduces non-linearity within the dynamics and generates measurement induced noisy dynamics. Established optimal control techniques have worked well for linear, unitary and deterministic systems, however no known generalized method exists for non-linear and stochastic systems. Reinforcement learning can be implemented in these settings because it is agnostic to the underlying physical description generating the observed dynamics. Control schemes can be derived heuristically using agent-based learning in a quantum environment. Such an approach would be adaptable and robust to changes in the environment and could be implemented more readily in experimental settings than model-based optimal control techniques. Beyond this, reinforcement learning has the potential to become a powerful simulation tool for quantum systems which cannot be analysed effectively by established methods such as systems with large Hilbert spaces, non-integrable systems and systems undergoing far from equilibrium dynamics. Learning-based approaches show promise as a way to probe the control landscapes of these quantum systems to

Modal Title

Ryan O’Connor

Student

Project Title

Leveraging machine learning for the design and analysis of optimisation algorithms

Project Description

Combinatorial optimization problems arise in many areas of computer science and other disciplines, such as business analytics, artificial intelligence and operations research. These problems typically involve finding groupings, orderings or assignments of discrete, finite sets of elements that satisfy certain conditions or constraints. Designing good optimization algorithms requires human ingenuity and a spark of genius. Automating the process of designing and analyzing algorithms has been a long-standing quest for AI researchers and such techniques are expected to have a very high impact in a range of applications. This PhD thesis will explore if machine learning techniques (particularly reinforcement learning and graph neural networks) can be leveraged to augment the human ability to design good heuristics for given input distributions. The thesis will also explore if reinforcement learning techniques can assist in finding counterexamples to discover the limitations of the existing heuristics in terms of the best approximation ratio achievable or in terms of running time. The success of this project will greatly augment the human ability to design algorithms for combinatorial optimization problems in industry. 39

Modal Title

Sonal Baberwal

Student

Project Title

Machine Learning Algorithms for exoskeleton control using Brain Computer Interfaces

Project Description

A Brain-Computer Interface (BCI) system allows the brain to communicate with an external device. A BCI consists of signal acquisition, feature extraction, feature translation, and device output. BCI systems have been a great help in the medical field for assistance to patients suffering from neuromotor disorders, spinal cord injuries and trauma to the nervous systems, with applications including wheelchairs and exoskeletons. Brain activity may be recorded through invasive, semi-invasive or noninvasive systems, detecting electrical or optical signals from the brain related to brain activity. Once the raw signals are collected from these systems, it is important to analyze and translate these signals to control an interfacing device in real-time. Hence, a robust framework is a requirement for any BCI system. Machine learning/ Deep Learning algorithms are now used for the processing of these signals and later translating them into action. The algorithms for BCI control need to be trained and tested. Advancement in the BCI systems has developed from moving a pointer cursor to operating a wheelchair using the brain commands. Various neuronal potentials may be used to control such commands, for instance, SSVEP, P300 and motor imagery signals. This study aims to evaluate the EEG signals from a non-invasive cap and translating these for a lower limb exoskeleton allowing dynamic balance without the use of crutches while walking. One of the major challenges faced in this field is the time investment for training purposes. The research will apply qualitative and quantitative analysis to evaluate the present state-of-art in BCI systems to operate exoskeletons.The goal is to design a BCI framework having minimal training and evaluation of various neuronal potentials that would decrease the training time for exoskeleton addressing the following possible research questions: Q1. Which neuronal potentials have the highest accuracy, efficiency and require less training? Q2. Do hybrid systems perform better as compared to traditional systems for applications like exoskeletons? Q3. Which algorithms hold higher accuracy and could be operated in real-time to minimize the training. Following on from these questions a framework will be generated for the advancement in locomotion for exoskeletons.

Modal Title

Suriyadeepan Ramamoorthy

Student

Project Title

Machine Learning for Precision Oncology

Project Description

Cancer treatment is unique as no single cancer is the same and the way it affects varies from person to person. Cancer has more than 20,000 known pathways driven by gene mutations. These must be identified and targeted with focused therapies rather than ad hoc generalized treatments. Standard cancer treatment is ineffective in >75% of the patients. Precision medicine is an effective alternative to standard chemotherapy for many patients. Prescribing a personalized combination of treatments for cancer patients using aggregated information from the patient’s molecular profile (-omics data), radiomics data, tumor tissue samples, and the patient’s medical history (EHR) is a challenging problem. Machine Learning holds the key to addressing this problem efficiently and accurately. With the advent of Deep Neural Networks, Machine Learning algorithms are now capable of dealing with the challenges of scalability and high dimensionality of data. Multi-modal Learning in Deep Neural Networks can integrate different types of data (spatial, sequential, time-series, tabular, etc,) from multiple sources to perform a prediction. These features indicate deep learning’s ability to transform big data from genomics, radiomics, histopathology, and EHR, into clinically actionable knowledge. Translation of AI to clinical environments could usher in a new era of Precision Oncology in which all the available treatments for cancer, including molecularly-targeted, immunotherapy, and cytotoxic chemotherapies, will be utilized effectively in a patient-specific manner. Once deployed, the oncologists will be able to prescribe treatments to patients using insights provided by a computer-based decision support system that selects a combination of treatments by maximizing potential therapeutic efficacy while minimizing the side effects associated with ineffective treatments. One of the major hurdles for successful translation of deep learning algorithms from research to practice in precision medicine is their interpretability to physicians. The current state-of-the-art techniques in drug-sensitivity prediction, cancer patient stratification, etc, face non-trivial challenges in terms of interpretability. The interpretation and understanding of how a complex model arrives at its decisions, referred to as a “black box” problem, is a significant roadblock preventing widespread adoption of machine learning-powered applications in the healthcare sector.

Modal Title

Tendai Mukande

Student

Project Title

Representation Learning for Cross-Domain Recommendation with Deep Neural Networks

Project Description

The extraordinary amount of daily generated content by users represents an invaluable source of information to predict their intention, likes, and dislikes. This heterogeneous information comes in multiple forms, from simple preferences such as likes and ratings to texts, social interactions, videos, and images. Recommender systems are an effective solution to overcome information overload and have been indispensable in various information access systems as well facilitating decision-making processes. Usually, recommender systems focus on just one of these sources of information. While these diverse contents often contribute to complementary views on the user preferences, their combination under a unifying framework promises to provide a comprehensive and timely representation of the user. Deep learning architectures offer a versatile environment to combine such heterogeneous information in a principled way for predictive tasks which is essential as the recommendation performance depends on the ability to describe complex user-item relations and preference. This research project aims at defining a Deep Learning architecture for combining this variety of sources, transfer and adapt knowledge across multiple domains. The aspects to be looked into include long-term prediction, dynamic sequential decision-making, and resolving the dimensionality problem, particularly for complex systems, adaptation to new situations as well as the proposed model optimization for long term recommendation accuracy. A novel interactive Deep Reinforcement Learning (DRL) based recommendation system will be studied, as well as interactions between the system and environment simulated by Recurrent Neural Networks (RNN).

Modal Title

Tlamelo Makati

Student

Project Title

Project Title Web Accessibility Using Machine Learning Tools

Project Description

The World Wide Web has become a ubiquitous platform for all kinds of everyday activities such as retail, entertainment, communication work, and education. The usability and accessibility of the web impact users of these services to varying degrees causing irritation to some and outright exclusion to others. People with disabilities and users of Assistive ICT such as screen readers, Switch devices and alternative I/O are particularly impacted by inaccessible websites. Access to Banking, Government Services and other important content and services is severely restricted. The response is in implementing Accessibility Standards and Guidelines such as the Web Content Accessibility Guidelines 2.1[1] and European Standards which embrace these such as the harmonised European standard[2]. These guidelines are organised around the principles of Perceivable, Operable Understandable and Robust content. These are referred to as the P.O.U.R principles. They specify requirements such as Alternative Text for images and captioning for videos. They also insist on operational requirements such as good navigation and the ability to turn off animations and blinking effects. Recent work has looked at the role of how AI and ML can help meet these guidelines. This has produced tools for creating alternative text for images and auto-captioning for videos. ML-driven processes such as text simplification and word prediction are useful utilities. Technologies such as Speech recognition can provide alternative input mechanisms. These techniques have been aggregated into intelligent overlays with varying degrees of success[3]. There are, however, problems with current versions of these technologies. They don’t work or only partially work in many instances causing major frustrations. There are ML approaches in other applications that the Web uses that could play a role here. Of particular interest is how ML can optimise navigation through websites. Work done in games for example could inform this work here. The question that this project addresses is what useful role can ML and AI play in making websites more accessible. Where the ML tools can best be deployed in development life-cycles. How they can best be supported through strategies like co-design and active inclusion. Of particular insight to this work is how intelligent web analytics and data insights from other web disciplines such as SEO and behaviour analysis can feed into the improvement of ML and AI web accessibility technologies.
Modal Title

Valerio Antonini

Student

Project Title

Graph Analytics for Spatio-Temporal data

Project Description

Large amounts of spatio-temporal data are increasingly collected and analysed in several domains, including neuroscience, epidemiology, transportation, social sciences, health, climate science and Earth sciences. Spatio-temporal data differ from relational data since the spatial and temporal attribute becomes a crucial feature which must be properly exploited, making the task even more demanding. Due to the presence of these two coordinates, the observations are not independent or identically distributed. The samples can be related or linked in some spatial regions of specific temporal moments. An additional challenge is the dynamic state of the observations: they can change properties or class depending on time and space. The traditional machine learning models are not properly suitable for mining patterns from these relations, leading to poor performances and misleading interpretation. Thus, the emerging field of spatio-temporal data mining (STDM) proposes new methods for addressing the analysis of event-based data. In order to take full advantage of the relations emerging in time and space, the most accurate way to process these data is by designing them as a graph. A graph is a network of nodes and edges which can have different weights defining the importance and type of the relation. The first challenge is to find the most accurate construction of the graph to represent patterns among data. Neighbourhood graph construction is crucial for the quality of the analysis in graph-based methods. This task can be addressed according to several techniques, each of them having own strengths and weaknesses. In the following stages other challenges can be community detection, nodes centrality, clustering, anomaly detection, frequent pattern mining, relationship mining, change detection, predictive learning and links prediction. The research seeks to find new efficient methods to collect, represent and process spatio temporal data by using the combination of graph-based techniques with unsupervised, supervised and semi-supervised machine learning.

Modal Title

Yingjie Niu

Student

Project Title

AI-first Finance: Discovering and Forecasting with Alternative Data

Project Description

In the financial area, decision-making traditionally relied on quantitative indicators collected from financial statements manually. In the last decade, the explosion in the sheer magnitude of data, such as financial news, earnings conference call voice recordings, SEC 10k reports, etc. has brought a huge opportunity and has been playing an increasing role in asset management, decision-making tasks. Each type of data has its own advantages and disadvantages. High-frequency textual data such as social blogs are relatively short and can reflect real-time events, but always involves a lot of noise. Medium-frequency text data including financial news usually have clean content because they are from the official provider, which makes it easy to process and analyze, but the information carried has a certain lag. The professional financial documents, such as 10k reports, are more reliable and contain large valuable information, but are released quarterly or annually. The aim of the project is to leverage the advantages of different types of data and modern natural language processing (NLP) and artificial intelligence (AI) technologies to make a precise financial market prediction and assist the investor’s decision-making process. Objective1: Develop an approach incorporating multi-source text data, i.e. low-frequency, medium-frequency, and high-frequency financial text sources, into financial prediction. Objective2: By building a Graph Convolutional Network (GCN) for economic entities, extract the relationship of different entities, and prompt the use of indirectly correlated text data. 40