Abstract: With a growing trend of employing machine learning algorithms to assist decision making, it is vital to inspect both the data and machine learning models for potential systematic deviations to achieve a trustworthy AI application. Detection of anomalous samples is a field of active research that aims to identify observations (a subgroup of samples) in a given data that deviate from some concept of normality. Its application is crucial across different domains, e.g., to understand data quality, mis-annotations, detecting adversarial attacks, monitoring model performance, and informing new data collection design. Using healthcare as a use case, data-centric techniques will be demonstrated to address specific questions, such as vulnerable groups, heterogeneous intervention effects and new class detection. Moreover, scientific discovery is being facilitated using generative models recently. However, principled evaluation of these models, in domain-agnostic and interpretable ways, is beneficial to efficiently exploit the unique generation capabilities of models. Beyond curated datasets that are often utilized to train machine learning models, data-centric analysis should also extend to traditional data sources, such as textbooks, to identify potential representation biases.

Abstract: Just like any quantum computing device, quantum annealers suffer from noise errors which limit their performance. An example of such errors is control errors which alter the coefficients of the problem Hamiltonian which leads to a perfectly operating quantum annealer to produce correct results to the wrong Hamiltonian. As a solution, additional error correction steps are implemented however error correction uses up too much time and space hence the motivation to introduce error mitigation strategies.

Abstract: Kitaev models are the cornerstone of topological quantum computing which is currently the best way to perform fault-tolerant quantum computation at the hardware level. This presentation will be focused on the introduction of different Kitaev models and show in each case their error correction properties.

Abstract: Pension schemes play a vital role in social protection and economic growth. However, pension uptake in Kenya has been low, resulting in non-inclusive social protection that leaves a major part of the population at the risk of financial hardship upon retirement. This is exacerbated by the increasing dependency ratio due to declining mortality and rising life expectancy. The aim of this study was to find a model that robustly predicts the uptake of pension among individuals from recent data for timely intervention. The 2018 Kenya FinAccess Household Survey data was utilized in the analysis. Tree-based machine learning models and logistic regression classifiers were compared on their ability to predict the pension uptake using socio-economic and demographic-related covariates and the optimal model used in the prediction. Ensemble tree-based models showed better metric performance than standalone classifiers. Moreover, Random Forest had the most optimal performance as compared to other tree-based models and Logistic Regression classifiers, with the area under the receiver operating characteristic curve being highest for both the down-sampled set and up-sampled data set. All features used in the final analysis had non-zero importance in predicting pension uptake. However, the most important factors were NHIF usage, monthly income, and bank usage. Random Forest is the most robust model, is, therefore, the most suitable model for pension uptake prediction. Further, high levels of education were associated with higher pension uptake. Financial education is recommended to improve pension uptake. As a strategy, a collaboration between NHIF, banks, and pension scheme providers is suggested to increase pension uptake.

Abstract: One of the major challenges of machine learning and obstacle to more general intelligence is so-called continual learning. This refers to the regime of learning tasks in sequence. Humans are very good continual learners. We can learn X and then learn Y without forgetting X, and in some cases also transfer learned aspects of X to Y. On the other hand artificial neural networks suffer from catastrophic forgetting, where information relevant to previously learned tasks is overwritten by the information required to effectively learn later tasks. In this talk I will discuss recent work trying to understand how the similarity of tasks in a continual learning setting affects the amount of forgetting observed in artificial networks.

Abstract: The current study set out to identify factors affecting the number of visitors to national parks using machine learning techniques. The results of different linear regression and random forest models on both the train and test sets were compared using RMSE and R². Taken together, both random forest and linear regression models were able to predict better on the train set, but all failed to make better predictions on the test set. Both linear regression and random forest models performed better using data from Akagera National Park than Volcanoes and Nyungwe National Parks. The most important features to explain the number of visits to national parks include price, park specific characteristics, and different months of the year whose features tend to vary from one park to another. This implies that forecasting future visits to different national parks will not only allow policy makers and the park management to make effective planning and efficient allocation of resources, but will also provide valuable information to various people as they plan to visit various national parks.

Abstract: this paper aims at addressing some challenges that prevent monitoring, evaluation and learning systems from effectively supporting the adaptive management. Using a case study of monitoring the drinking water quality compliance in Rwanda, a proposed four-step approach illustrated how big data can be integrated and leveraged to guarantee accurate and timely data insights as well as reliable predictions and user-friendly reporting. That approach capitalized on Hadoop data lake and R/RStudio to ingest and store structured and unstructured data from different sources, process them, and finally to consume them into interactive dashboards created in Microsoft Power Business Intelligence. Compared to the traditional practice of considering only measurable indicators, more quality key performance indicators were introduced to support the acquisition of additional and triangulated evidence on performance, which should satisfy the information needs of stakeholders.

Abstract: Infection of cesarean section wounds is a leading cause of maternal morbidity in Rwanda. Early detection and care are essential to adequate treatment of surgical site infections (SSIs); however, there are often considerable barriers to receiving follow-up care. Our group has developed a machine learning (ML) SSI diagnostic algorithm to facilitate home-based SSI diagnosis, using either wound photographs or thermal images. In this talk, we will describe the development of these algorithms and present the robustness of the algorithms’ accuracy when the quality of the images is degraded .

Abstract: Stunting is still one of the most serious health and welfare issues globally. Therefore, we set out to apply machine learning (ML) classification algorithms on the Demographic Health Survey (DHS) data to predict stunting among under-five children in Zambia and to evaluate the prediction performance of each model. We applied Logistic regression (LR), Random Forest (RF), SV classification (SVC), XG Boost (XgB) and Naïve Bayes (NB) classification models to predict the probability of stunting among children under five years of age, using the 2018 DHS dataset with 6799 instances. All models were trained based on 3-fold cross-validation. We calibrated predicted probabilities and plotted the calibration curves to compare model performance. We computed accuracy, recall, precision, and F1 for each variable machine learning algorithm. Two thousand three hundred and twenty-seven (34.2%) of the children were stunted. 13 of 58 features were selected for inclusion in the model using random forest. Calibrating the predicted probabilities improved the performance of machine learning algorithms when evaluated using calibration curves. Random Forest was the most accurate algorithm with an accuracy score of 79% in the testing and 61.6% in the training data. The Random Forest classification model had the highest predictive accuracy for predicting stunting among children under five years of age using the ZDHS dataset. While Naïve Bayesian is the worst performing machine learning algorithm for predicting stunting using the ZDHS dataset. ML models aid the timely development of interventions aimed at preventing stunting.

Abstract: Actuaries utilize demographic features such as mortality and longevity rates for pricing, valuation, and reserving life insurance and pension contracts. Capturing accurate mortality estimates requires factual mortality assumptions in mortality models. However, the dynamic and uncertain nature of mortality improvements and deteriorations necessitates better approaches in tracking mortality changes, for instance, using the causes of deaths features. This research aims to determine temporal homogeneous clusters using unsupervised learning, clustering approach to group causes of death based on (dis)similarity measures to set representative clusters in detection and monitoring death trends. The causes of death dataset were derived from the World Health Organization, Global Health Estimates for males and females, from 2000 to 2019 for Kenya. A hierarchical agglomerative clustering technique was implemented with a modified Dynamic Time Warping distance criteria. Between six and 14 clusters were optimally achieved for both males and females. Using visualizations, principal clusters were detected. Over time, causes of death trends of these clusters have shown a correlated association with mortality and longevity rates and rationalize why insurance and pension offices may include this approach as a preliminary step to undertake mortality and longevity modelling.

Abstract: Mechanical engineering is a branch of Engineering which deals with mechanical principles applications in product design, development, manufacturing, and energy.

As society’s challenges become more complex as we chase NetZero, engineers are turning data science and machine learning to real-world problems, researchers need to understand how to work with nonlinear systems, in complex environments and with ever-changing factors.

Data science, machine learning and AI have become important teaching and research tools in managing and describing big data sets, finding patterns, and solving problems that might otherwise show up only after long phases of testing. This lecture highlights how Mechanical and Energy Engineering offers a promising avenue for data science applications.

Abstract: Networks are all around us. Complex interactions between entities play a profound role in society and the natural world, from the World Wide Web and molecular processes to food webs and economic networks. But few networks optimise the need for a greater understanding than financial networks. We are part of financial networks when we make payments, borrow money or invest in financial markets.

This high-level presentation will provide an introduction to network analytics and why it is quickly becoming a key tool for supervisors in attaining visibility over the systems they oversee. Dr Kimmo Soramäki will also discuss various real-world applications and case studies of network analytics and simulation in finance, banking and regulation.

Abstract: Some of the topics to be discussed will include;

• How to prepare a scientific talk?

• How to write a research paper? Route to publication

• How to manage supervisors, mentees, and research collaborations?

• How to navigate a conference?

• How to design a course?

• How to write a research proposal?

If you have additional questions you would like to discuss, please submit them here:

https://forms.gle/qg6KZxyUb8Uncwff9

Abstract: In 2020 there were an estimate of 149.2 million children under the age of five were reported to be stunted globally. In Rwanda, there is a significance decrease in the prevalence of stunting from 51% in 2005 to 33% in 2020. The effective stewardship of health programs requires precise targeting and intervention for the needs of the population. Geospatial survey data analysis can better guide program to improve health and development outcomes. The objectives of this study are to estimate the prevalence and distribution of under-five stunted children at cluster level in Rwanda as well as making predictions and provides the exceedance probability maps for each of stunting at a threshold value of 30%. This study uses data from the Rwandan Demographic and Health Surveys (R-DHS) 2014/2015. Bayesian model is implemented using integrated nested Laplace approximations (INLAs) algorithm to generate approximations of the marginal posterior distribution of the stunting prevalence at each location across Rwanda. As revealed in this study, the prevalence of stunting was found at 420 clusters. The lowest prevalence was seen in clusters of Kigali city and ranged between 0-25 percent. The highest prevalence was seen in the Western Province specifically in Congo Nile Divide. Stunting prevalence exceeding a 30% probability in Rwanda, allowing for the identification of the locations and communities most at risk of stunting. In Rwanda, regional disparities in childhood stunting are significant. More focus should be oriented in the western province and in Congo Nile Divide Zone. The maps also show areas when the prevalence of stunting exceeds pre-determined thresholds. They can be used as a useful tool for program managers and implementers.

Abstract: Biodiversity is declining globally at unprecedented rates. We need to monitor species in real time and in greater detail to quickly understand which conservation efforts are most effective and take corrective action. Current ecological monitoring systems generate data far faster than researchers can analyze it, making scaling up impossible without automated data processing. However, ecological data collected in the field presents a number of challenges that current methods, like deep learning, are not designed to handle. We make biodiversity-inspired methodological improvements to machine learning systems to help make efficient, scalable biodiversity monitoring systems a reality. In parallel to this work, we provide off-the-shelf AI-assisted tools to ecologists that allow them to efficiently process their data and focus their efforts on analyzing the most challenging cases.

Abstract: In this talk, I will give an overview of some past and current projects that I am working on (I have worked on). The main presentation will focus on adaptive learning from a military and civil aviation perspective. More specifically, the use of AI-Machine Learning to improve and maintain US Air Force pilots' knowledge at a certain required level throughout their services. I will also briefly talk about some of the applications of AI-Machine Learning in the banking system.

Abstract: Data is widely collected and stored using data models such as entity relationships or relational models to create databases. A particular database is a Knowledge Base (KB). A knowledge base (KB) aims to accumulate and convey knowledge of the real world. KBs that are stored using graph-structured data models are called Knowledge Graphs (KGs). KGs are in general incomplete though they may contain millions of relational facts. If the underlying structure of a KG is known, it can then be used to complete the KB or to find out if there is a relation between two entities of interest. This is typically referred to as link prediction tasts. Machine learning (ML) techniques have great potential to tackle these tasks. However, ML techniques can not be directly used to analyze KBs since they are symbolic and logic databases. One way to circumvent this challenge is to project the KG into a continuous vector space.

In recent works, several embedding models have been proposed. The goal of this talk is to introduce the concept of Knowledge Graph Embeddings (KGE). We will construct from scratch a small KG and study some of the state-of-the-art translational embeddings by describing them and explaining the geometric intuition behind them.

Abstract: A newborn’s health is a primary factor that determines the overall health of a human being and even its life expectancy. Therefore, its health should be monitored not only after birth but also when the baby is still growing in the womb. Birth weight is one of the crucial aspects to be observed. Low birth weight is among the main problems that newborns face. Low birth weight (LBW) is the weight at birth less than 2500g as defined by the World Health Organization. A global estimate of 15 to 20 percent of total live births are low birth weight representing over twenty million births every year. In Kenya, the rate of children born with low weight is 8 percent. This rate is still alarming. Several methods have been used to measure and approximate birth weight in clinical practice including obstetric ultrasound, symphysio-fundal height measurements and abdominal palpation. However, these methods are associated with reliability and accuracy challenges therefore, calling for more robust methods. This research aims at using supervised machine learning techniques for predicting

low birth weight using the maternal risk factors that have been proven to be associated with low birth weight.

Abstract: The advancement and affordability of hardware has made it possible for machine learning research to be applied to real-world problems. However, the application of the research has given rise to new challenges given the environment which the research was conducted and that which the research is applied is different. There is a call by researchers to think about the effects of their research to the real-world: effects on the environment, societal biases and fairness, scalability, usability, among others. With the push of African continent towards 4 th Industrial revolution, researchers need to think about how their research will be applied in the continent. In this talk, we talk about what it takes to move research to production, some of the challenges we encountered and finally the complexities of moving research to production in the African continent and the opportunities that come with them.

Abstract: With the advance of technology, the tax base has also become wider and as a result, the tax fraud is growing. Depending on the type of dataset used, fraud detection experts and researchers have used different methods to identify questionable cases. This paper aims to predict features of tax fraud using the best supervised machine learning model. This research provides a context in which a machine learning model can be used by the fraud expert, and an implemented model provides instant feedback to the fraud expert. Different supervised machine learning models were evaluated. Logistic regression was the most robust model in predicting tax fraud. Findings revealed that taxpayers who import and export goods, taxpayers with domestic businesses, those with no losses, taxpayers whose businesses are under services sectors are highly related to tax fraud. This research is one of the first to compare the performance of various supervised models of machine learning with a real dataset to identify tax fraud factors.

Abstract: here

Abstract: This study estimates the causal effect of Rwanda’s unconditional cash transfer program (VUP-Direct Support) on the incidence of poverty, the poverty gap, and household food and non-food expenditure for direct support recipients. Our empirical analysis applies four matching methods to data from the 2013/14 household survey in order to estimate the program impact on the treated. The findings show that participation in the program has positive and statistically significant effects on measured headcount poverty and poverty gap. The program results in an increase in both total and food consumption, with a reduction in consumption of food from home production, and no change in non-food consumption. The fact that average annual cash transfers are equivalent to a third of total consumption, for recipients, plays an essential role in the observed results. The most important finding may be that in measuring the impact of direct transfers, it is essential to allow for behavioral responses, such as reducing auto-consumption, or increasing investment in assets and farming. A purely accounting approach to the impact of direct transfers seriously overstates the effects on poverty and consumption.

Keywords: Cash transfers, consumption, poverty, propensity score, Rwanda.

Abstract: The objective of this paper was to analyze the topology of the interbank network in Rwanda for policy formulation. Our main result is that the interbank market network in Rwanda is described by a core- periphery model with some level of completeness of the interbank market in Rwanda. As policy implication, any risk from a bank is more easily shared within the interbank market network provided that there is nearly a complete network. This is an indication that the risk of instability of the financial system in Rwanda originating from the interbank market is limited.

Abstract: The phenomenon of topological phase of matter in quantum physics is studied

using artificial neural networks. This is demonstrated by using the Kitaev toric code states

with intrinsic topological order – a long range property in the ground state. We will show

how the topological ground states of the toric code can be represented by short range neural

networks in an exact and efficient fashion – that is the required number of hidden neurons is as small as the number of physical spins and the number of parameters scales only linearly

with the system size. In addition, we will show that the proposed short-range neural networks can describe the excited states with abelian anyons and their non-trivial mutual statistics as

well.

Abstract: Quantum-access security, where an attacker is granted superposition access to secret-keyed functionalities, is a fundamental security model and its study has inspired results in post-quantum security. Following the technique of security proposed by Alagic et al.~(Eurocrypt 2020), we analyse the quantum-access security for the Winternitz OTS (W-OTS) in the Quantum Random Oracle Model (QROM). First, we develop a tool for the analysis of hash chains in the Quantum Random Oracle Model based on the superposition oracle technique by Zhandry (Crypto 2019) that we use as keys in our analysis and might be of independent interest. Next, we study the security of the W-OTS in situations where i) a quantum adversary only has access to the signing oracle function of the W-OTS. ii) in addition to having access to the signing oracle, the adversary has access to a random oracle, which could help him to get information about the keys. Then use it to produce a fresh forgery. Finally, we show that any quantum adversary only has a small probability to break the security of the W-OTS.

Abstract: By collecting sufficient data and turning them into actionable insights, an effective MEL system supports the adaptive management, a necessary process to ensure a successful achievement of program results. Integrating big data can help address challenges such as inaccurate performance measures, inability to make reliable predictions, delayed implementation of corrective actions, and inability of some users to understand and use MEL reports. A six-step exploratory approach using Hadoop ecosystem can help integrate big data collected on key performance indicators (KPIs). Accordingly, the data primarily stored in Hadoop data lake can be processed through Apache Spark (in-memory processing) and/or Spark cluster in Azure HDInsight. Those systems allow the use of machine learning and can be accessed by the Microsoft Power business intelligence (BI) to create interactive dashboards that can map interventions, perform calculations, assess the performance, identify trends, and provide real-time information on KPIs. As a result, MEL systems will enjoy extended opportunities to take advantage on huge amount of big data to draw reliable data insights in a more user-friendly reporting platform.

Abstract: In many parts of the world which are dependent on small scale, rain fed farming, historical rainfall data is essential in providing useful climate information for agricultural decision making. Climate impacts are local, and therefore locally relevant data is required.

Traditionally, this data has come from ground stations. This presents a challenge in many

Africa countries, where the network of ground stations do not provide adequate spatial

coverage, particularly away from major towns and cities. Satellite products are increasingly

being used instead and potentially address some of these issues due to their high spatial

and temporal resolution and global or continent wide coverage. In this talk we present a

methodology for comparison of gridded rainfall products with station data with example

data from Africa. Differences in the data sources are shown and potential bias correction

methods are discussed. Finally, we also describe the current methods used to produce

satellite rainfall estimates, based on cloud information, and discuss potential for improving

these by making more use of the satellite imagery.

Abstract: Smart Grid uses modern metering electricity and some devices that collect

energy data in a real time manner and send the clients usage report of electricity usage to

the service provider. The service provider uses the received data for billing the client or

other services. Through this smart grid, the daily energy consumption and devices that are

being used by home owner can be predicted by the service provider depending on how

electricity is consumed. This can lead to security issues security where hackers can easily

capture clients’ data while it’s being transferred to the service provider. The hacker can

modify the transmitted data and the services provider will receive the wrong data. This

paper introduces a key distribution system that is more efficient and secure. Existing

identity based encryption and identity based signature schemes for smart grid have key

escrow problem. In this paper we introduce a certificateless signcryption for key distribution

scheme which is more efficient and secure than the existing schemes. It allows for bothdecryption and verification by authorized users, provide Key Generation Center to only

partial key and provide low computation and communication cost compared with existing

works. The proposed scheme also achieves key escrow resilience unlike previous works in

this field

Abstract: here

Abstract:

Abstract: In Rwanda, the estimated out of pocket health expenditure has been increased from 24.46 % in 2000 to 26 % in 2015. Despite the existence of guidelines in estimation of out of pocket health expenditures provided by WHO (2018), the estimation of out of pocket health expenditure still has difficulties in many countries including Rwanda.

The purpose of this paper was to figure out the best model which predicts the out of pocket health expenditures in Rwanda during the process of considering various techniques of machine learning by using the Rwanda Integrated Living Conditions Surveys (EICV5) of 14580 households (2018).

Our findings presented the models which predict the out of pocket health expenditures with higher accuracy. Furthermore, machine learning techniques were used to judge which predictor variable was important in our prediction process and comparison of the performance of the algorithms through train accuracy and test accuracy metric measures. Finally, the findings show that the tests of accuracy of the models were 50.16 % for MARS model, 74 % Decision tree model, 87% for treenet model, 83% for Random forest model, Gradient boosting 81% predictor total consumption played a significant role in the model for all tested models.

Finally, we conclude that the total consumption of the household came out to be the most important variable which is consistently true to all the algorithms tested. The findings from our study have policy implications for policy makers in Rwanda and in the world generally. We recommend the government to significantly increase public spending on health. Domestic financial resources are key to moving closer to Universal health coverage (UHC) and should be increased on a long-term basis. In addition, these results will be useful for the future to assess the out of pocket health expenditures dataset.

Abstract: Heart failure (HF) has become a public health problem worldwide. However, most of the information available in the existing published literature about HF epidemiology and control is from high income countries. This research paper shows the distribution, mapping, summary clinical characteristics and New York Heart Association (NYHA) classification of HF as well as its main risk factors in Rwanda.

A study design is retrospective and a descriptive statistical analysis was performed to obtain the results. Secondary data collected from all available 4,085-HF files of patients that were hospitalized in seven hospitals in Rwanda from 2008 to 2019 were utilized.

The findings revealed an unequal distribution of HF patients across all 30 administrative districts of Rwanda, as 71.8 percent of HF patients resided in only 10 districts. The results further showed that the major symptoms of HF in Rwanda, were dyspnea representing 63.6 percent of patients, edema (35.6 percent), persistent cough (27.5 percent), and abdominal swelling (24.4 percent). Furthermore, dilated cardiomyopathy, valvular heart disease, hypertension (especially among adults), and congenital heart defect (common among young people under 18 years) were found to be the common risk factors for HF in Rwanda. This study found that the main risk factors for HF that showed disparity between young people under18 years old and adults were valvular heart diseases (52.7 percent for young people under 18years old against 41.5 percent for adults), congenital heart defects (32.4 percent for young people under 18years old against 1.5 percent for adults), hypertension (24.1 percent for adults against 7.8 percent for under 18years), and cardiomyopathy (44.5 percent for adults against 24.8 percent for under 18years). The research findings also indicated that about a half (48.9 percent) of the patients hospitalized for HF fell in NYHA class III (35.6 percent) and class IV (13.3 percent).

HF sufferers together with their families are disrupted by the burden of HF in Rwanda. Promoting awareness about HF condition, facilitating access to capable health facilities in all districts of Rwanda, diagnosing, treating and controlling the risk factors of HF at earlier stages, could significantly reduce the impact of HF on sufferers and their surroundings.

Abstract:

Abstract: Smart Grid uses modern metering electricity and some devices that collect energy data in a real time manner and send the clients usage report of electricity usage to the service provider. The service provider uses the received data for billing the client or other services. Through this smart grid, the daily energy consumption and devices that are being used by home owner can be predicted by the service provider depending on how electricity is consumed. This can lead to security issues security where hackers can easily capture clients data while it’s being transferred to the service provider. The hacker can modify the transmitted data and the services provider will receive the wrong data. This paper introduces a key distribution system that is more efficient and secure. Existing identity based encryption and identity based signature schemes for smart grid have key escrow problem. In this paper we introduce a certificateless signcryption for key distribution scheme which is more efficient and secure than the existing schemes. It allows for both decryption and verification by authorized users, provide Key Generation Center to only partial key and provide low computation and communication cost compared with existing works. The proposed scheme also achieves key escrow resilience unlike previous works in this field.

Keywords— Certificateless, Anonymous, signcryption, Smart Grid, Smart Meters, Key distribution

Abstract: Every year, 435,000 people worldwide die from Malaria, mainly in Africa and Asia. However, malaria is a curable and preventable disease. Most countries are developing malaria elimination plans to meet sustainable development goal three, target 3.3, which includes ending the epidemic of malaria by 2030. Rwanda, through the malaria strategic plan 2012-2018 set a target to reduce malaria incidence by 42% from 2012 to 2018. Assessing the health policy and taking a decision using the incidence rate approach is becoming more challenging. We are proposing suitable statistical methods that handle spatial structure and uncertainty on the relative risk that is relevant to National Malaria Control Program. We used a spatio-temporal model to estimate the excess probability for decision making at a small area on evaluating reduction of incidence. SIR and BYM models were developed using routine data from Health facilities for the period from 2012 to 2018 in Rwanda. The fitted model was used to generate relative risk (RR) estimates comparing the risk with the malaria risk in 2012, and to assess the probability of attaining the set target goal per area. The results showed an overall increase in malaria in 2013 to 2018 as compared to 2012. Ofall sectors in Rwanda, 47.36% failed to meet targeted reduction in incidence from 2012 to 2018. Our approach of using excess probability method to evaluate attainment of target or identifying threshold is a relevant statistical method, which will enable the Rwandan Government to sustain malaria control and monitor the effectiveness of targeted interventions.

Abstract: Graphical semi-supervised learning is the problem of labeling the vertices of a graph given the labels of a few vertices along with geometric information about the graph. Such problems have attracted a lot of attention in machine learning for classification of large data sets. In this talk we give an introduction to graph-based learning, and in particular, semi-supervised learning techniques, with a focus on recent research.