Carnegie Mellon University Africa

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of dailylife activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards, with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception. Project page: https://ego4d-data.org/

Sentiment analysis has become an important area of research in natural language processing. This technique has a wide range of applications, such as comprehending user preferences in ecommerce feedback portals, politics, and in governance. However, accurate sentiment analysis requires robust text representation techniques that can convert words into precise vectors that represent the input text. There are two categories of text representation techniques: lexicon-based techniques and machine learning-based techniques. From research, both techniques have limitations. For instance, pre-trained word embeddings, such as Word2Vec, Glove, and bidirectional encoder representations from transformers (BERT), generate vectors by considering word distances, similarities, and occurrences ignoring other aspects such as word sentiment orientation. Aiming at such limitations, this paper presents a sentiment classification model (named LeBERT) combining sentiment lexicon, N-grams, BERT, and CNN. In the model, sentiment lexicon, N-grams, and BERT are used to vectorize words selected from a section of the input text. CNN is used as the deep neural network classifier for feature mapping and giving the output sentiment class. The proposed model is evaluated on three public datasets, namely, Amazon products’ reviews, Imbd movies’ reviews, and Yelp restaurants’ reviews datasets. Accuracy, precision, and F-measure are used as the model performance metrics. The experimental results indicate that the proposed LeBERT model outperforms the existing state-of-the-art models, with a F-measure score of 88.73% in binary sentiment classification.

Abstract We incorporate communication into the multi-UAV path planning problem for search and rescue missions to enable dynamic task allocation via information dissemination. Communication is not treated as a constraint but a mission goal. While achieving this goal, our aim is to avoid compromising the area coverage goal and the overall mission time. We define the mission tasks as: search, inform, and monitor at the best possible link quality. Building on our centralized simultaneous inform and connect (SIC) path planning strategy, we propose two adaptive strategies: (1) SIC with QoS (SICQ): optimizes search, inform, and monitor tasks simultaneously and (2) SIC following QoS (SIC+): first optimizes search and inform tasks together and then finds the optimum positions for monitoring. Both strategies utilize information as soon as it becomes available to determine UAV tasks. The strategies can be tuned to prioritize certain tasks in relation to others. We illustrate that more tasks can be performed in the given mission time by efficient incorporation of communication in the path design. We also observe that the quality of the resultant paths improves in terms of connectivity.

For sustainability and efficiency in maintaining high crop yield and less chemically polluted agricultural lands, precise weed mapping is essential for the total implementation of site-specific weed management which currently stands as a major challenge in present day agriculture. In this research, the robustness of the training epochs of You Only Look Once (YOLO) v5s, a Convolutional Neural Network (CNN) model was evaluated for the development of an automatic crop and weeds classification using UAV images. The images were annotated using a bounding box and they were trained on google colaboratory over 100, 300, 500, 600, 700 and 1000 epochs. The model detected and categorized five different classes which are sugarcane (Saccharum officinarum), banana trees (Musa), spinach (Spinacia oleracea), pepper (Capsicum), and weeds. To find the optimal performance on the test set, the model was trained across several epochs, and training was stopped when the test performance (classification accuracy, precision, and recall) began to drop. The obtained result shows that the performance of the classifier improved significantly as the range of training epochs tends to rise from 100 through to 600 epochs. Meanwhile, a slight decline was observed as the number of epoch was increased to 700 when the classification accuracy, the precision of weed and recall of 65, 43 and 43%, respectively, was recorded as against 67, 78 and 34% that was obtained as the classification accuracy, weed precision and recall, respectively, at 600 epochs. This decline continued even when the epoch was increased to 1000 where classification accuracy, weed precision and recall of 65%, 45% and 40%, respectively was obtained. The results showed that the training epoch of YOLOv5s significantly affects the model's robustness in automatic crop and weep classification and identified 600 as the epoch for optimal performance.

Robot-assisted therapy (RAT) offers potential advantages for improving the social skills of children with autism spectrum disorders (ASDs). This article provides an overview of the developed technology and clinical results of the EC-FP7-funded Development of Robot-Enhanced therapy for children with AutisM spectrum disorders (DREAM) project, which aims to develop the next level of RAT in both clinical and technological perspectives, commonly referred to as robot-enhanced therapy (RET). Within this project, a supervised autonomous robotic system is collaboratively developed by an interdisciplinary consortium including psychotherapists, cognitive scientists, roboticists, computer scientists, and ethicists, which allows robot control to exceed classical remote control methods, e.g., Wizard of Oz (WoZ), while ensuring safe and ethical robot behavior. Rigorous clinical studies are conducted to validate the efficacy of RET. Current results indicate that RET can obtain an equivalent performance compared to that of human standard therapy for children with ASDs. We also discuss the next steps of developing RET robotic systems.

Coronavirus disease (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), is rapidly spreading across the globe. The clinical spectrum of SARS-CoV-2 pneumonia requires early detection and monitoring, within a clinical environment for critical cases and remotely for mild cases, with a large spectrum of symptoms. The fear of contamination in clinical environments has led to a dramatic reduction in on-site referrals for routine care. There has also been a perceived need to continuously monitor non-severe COVID-19 patients, either from their quarantine site at home, or dedicated quarantine locations (e.g. hotels). In particular, facilitating contact tracing with proximity and location tracing apps was adopted in many countries very rapidly. Thus, the pandemic has driven incentives to innovate and enhance or create new routes for providing healthcare services at distance. In particular, this has created a dramatic impetus to find innovative ways to remotely and effectively monitor patient health status. In this paper, we present a review of remote health monitoring initiatives taken in 20 states during the time of the pandemic. We emphasize in the discussion particular aspects that are common ground for the reviewed states, in particular the future impact of the pandemic on remote health monitoring and consideration on data privacy.

AIM: HIV prevention measures in sub-Saharan Africa are still short of attaining the UNAIDS 90-90-90 fast track targets set in 2014. Identifying predictors for HIV status may facilitate targeted screening interventions that improve health care. We aimed at identifying HIV predictors as well as predicting persons at high risk of the infection. METHOD: We applied machine learning approaches for building models using population-based HIV Impact Assessment (PHIA) data for 41,939 male and 45,105 female respondents with 30 and 40 variables respectively from four countries in sub-Saharan countries. We trained and validated the algorithms on 80% of the data and tested on the remaining 20% where we rotated around the left-out country. An algorithm with the best mean f1 score was retained and trained on the most predictive variables. We used the model to identify people living with HIV and individuals with a higher likelihood of contracting the disease. RESULTS: Application of XGBoost algorithm appeared to significantly improve identification of HIV positivity over the other five algorithms by f1 scoring mean of 90% and 92% for males and females respectively. Amongst the eight most predictor features in both sexes were: age, relationship with family head, the highest level of education, highest grade at that school level, work for payment, avoiding pregnancy, age at the first experience of sex, and wealth quintile. Model performance using these variables increased significantly compared to having all the variables included. We identified five males and 19 females individuals that would require testing to find one HIV positive individual. We also predicted that 4·14% of males and 10.81% of females are at high risk of infection. CONCLUSION: Our findings provide a potential use of the XGBoost algorithm with socio-behavioural-driven data at substantially identifying HIV predictors and predicting individuals at high risk of infection for targeted screening.

In this work we present a fast kinodynamic RRT-planner that uses dynamic nonprehensile actions to rearrange cluttered environments. In contrast to many previous works, the presented planner is not restricted to quasi-static interactions and monotonicity. Instead the results of dynamic robot actions are predicted using a black box physics model. Given a general set of primitive actions and a physics model, the planner randomly explores the configuration space of the environment to find a sequence of actions that transform the environment into some goal configuration. In contrast to a naive kinodynamic RRT-planner we show that we can exploit the physical fact that in an environment with friction any object eventually comes to rest. This allows a search on the configuration space rather than the state space, reducing the dimension of the search space by a factor of two without restricting us to non-dynamic interactions. We compare our algorithm against a naive kinodynamic RRT-planner and show that on a variety of environments we can achieve a higher planning success rate given a restricted time budget for planning.

Governments implemented many non-pharmaceutical interventions (NPIs) to suppress the spread of COVID-19 with varying results. In this paper, country-level daily time series from Our World in Data facilitates a global analysis of the propagation of the virus, policy responses and human mobility patterns. High death counts and mortality ratios influence policy compliance levels. Evidence of long-term fatigue was found with compliance dropping from over 85% in the first half of 2020 to less than 40% at the start of 2021, driven by factors such as economic necessity and optimism coinciding with vaccine effectiveness. NPIs ranged from facial coverings to restrictions on mobility, and these are compared using an empirical assessment of their impact on the growth rate of case numbers. Masks are the most cost-effective NPI currently available, delivering four times more impact than school closures, and approximately double that of other mobility restrictions. Gathering restrictions were the second most effective. International travel controls and public information campaigns had negligible effects. Literacy rates and income support played key roles in maintaining compliance. A 10% increase in literacy rate was associated with a 3.2% increase in compliance, while income support of greater than half of previous earnings increased compliance by 4.8%.

This paper considers the problem of image set-based face verification and identification. Unlike traditional single sample (an image or a video) setting, this situation assumes the availability of a set of heterogeneous collection of orderless images and videos. The samples can be taken at different check points, different identity documents $etc$ . The importance of each image is usually considered either equal or based on a quality assessment of that image independent of other images and/or videos in that image set. How to model the relationship of orderless images within a set remains a challenge. We address this problem by formulating it as a Markov Decision Process (MDP) in a latent space. Specifically, we first propose a dependency-aware attention control (DAC) network, which uses actor-critic reinforcement learning for attention decision of each image to exploit the correlations among the unordered images. An off-policy experience replay is introduced to speed up the learning process. Moreover, the DAC is combined with a temporal model for videos using divide and conquer strategies. We also introduce a pose-guided representation (PGR) scheme that can further boost the performance at extreme poses. We propose a parameter-free PGR without the need for training as well as a novel metric learning-based PGR for pose alignment without the need for pose detection in testing stage. Extensive evaluations on IJB-A/B/C, YTF, Celebrity-1000 datasets demonstrate that our method outperforms many state-of-art approaches on the set-based as well as video-based face recognition databases.

Abstract Social media has been embraced by different people as a convenient and official medium of communication. People write or share messages and attach images and videos on Twitter, Facebook and other social media platforms. It therefore generates a lot of data that is rich in sentiments. Sentiment analysis has been used to determine the opinions of clients, for instance, relating to a particular product or company. Lexicon and machine learning approaches are the strategies that have been used to analyze these sentiments. The performance of sentiment analysis is, however, distorted by noise, the curse of dimensionality, the data domains and the size of data used for training and testing. This article aims at developing a model for sentiment analysis of social media data in which dimensionality reduction and natural language processing with part of speech tagging are incorporated. The model is tested using Naïve Bayes, support vector machine, and K‐nearest neighbor algorithms, and its performance compared with that of two other sentiment analysis models. Experimental results show that the model improves sentiment analysis performance using machine learning techniques.

Significance While many forecasters are moving toward generating probabilistic predictions, energy forecasts typically still consist of point projections and scenarios without associated probabilities. Empirical density forecasting methods provide a probabilistic amendment to existing point forecasts. Here we lay the groundwork for evaluating the performance of these methods in the data-scarce setting of long-term forecasts. Results can give policy analysts and other users confidence in estimating forecast uncertainties with empirical methods.

Due to the massive adoption of mobile money in Sub-Saharan countries, the global transaction value of mobile money exceeded $2 billion in 2021. Projections show transaction values will exceed $3 billion by the end of 2022, and Sub-Saharan Africa contributes half of the daily transactions. SMS (Short Message Service) phishing cost corporations and individuals millions of dollars annually. Spammers use Smishing (SMS Phishing) messages to trick a mobile money user into sending electronic cash to an unintended mobile wallet. Though Smishing is an incarnation of phishing, they differ in the information available and attack strategy. As a result, detecting Smishing becomes difficult. Numerous models and techniques to detect Smishing attacks have been introduced for high-resource languages, yet few target low-resource languages such as Swahili. This study proposes a machine-learning based model to classify Swahili Smishing text messages targeting mobile money users. Experimental results show a hybrid model of Extratree classifier feature selection and Random Forest using TFIDF (Term Frequency Inverse Document Frequency) vectorization yields the best model with an accuracy score of 99.86%. Results are measured against a baseline Multinomial Naïve-Bayes model. In addition, comparison with a set of other classic classifiers is also done. The model returns the lowest false positive and false negative of 2 and 4, respectively, with a Log-Loss of 0.04. A Swahili dataset with 32259 messages is used for performance evaluation.

Denoising computed tomography (CT) medical images is crucial in preserving information and restoring images contaminated with noise. Standard filters have extensively been used for noise removal and fine details’ preservation. During the transmission of medical images, noise degrades the visibility of anatomical structures and subtle abnormalities, making it difficult for radiologists to accurately diagnose and interpret medical conditions. In recent studies, an optimum denoising filter using the wavelet threshold and deep-CNN was used to eliminate Gaussian noise in CT images using the image quality index (IQI) and peak signal-to-noise ratio (PSNR). Although the results were better than those with traditional techniques, the performance resulted in a loss of clarity and fine details’ preservation that rendered the CT images unsuitable. To address these challenges, this paper focuses on eliminating noise in CT scan images corrupted with additive Gaussian blur noise (AGBN) using an ensemble approach that integrates anisotropic Gaussian filter (AGF) and wavelet transform with a deep learning denoising convolutional neural network (DnCNN). First, the noisy image is denoised by AGF and Haar wavelet transform as preprocessing operations to eliminate AGBN. The DnCNN is then combined with AGF and wavelet for post-processing operation to eliminate the rest of the noises. Specifically, we used AGF due to its adaptability to edge orientation and directional information, which prevents blurring along edges for non-uniform noise distribution. Denoised images are evaluated using PSNR, mean squared error (MSE), and the structural similarity index measure (SSIM). Results revealed that the average PSNR value of the proposed ensemble approach is 28.28, and the average computational time is 0.01666 s. The implication is that the MSE between the original and reconstructed images is very low, implying that the image is restored correctly. Since the SSIM values are between 0 and 1.0, 1.0 perfectly matches the reconstructed image with the original image. In addition, the SSIM values at 1.0 or near 1.0 implicitly reveal a remarkable structural similarity between the denoised CT image and the original image. Compared to other techniques, the proposed ensemble approach has demonstrated exceptional performance in maintaining the quality of the image and fine details’ preservation.

In recent years, several factors such as environmental pollution, declining fossil fuel supplies, and product price volatility have led to most countries investing in renewable energy sources. In particular, the development of photovoltaic (PV) microgrids, which can be standalone, off-grid connected or grid-connected, is seen as one of the most viable solutions that could help developing countries such as Rwanda to minimize problems related to energy shortage. The country’s current electrification rate is estimated to be 59.7%, and hydropower remains Rwanda’s primary source of energy (with over 43.8% of its total energy supplies) despite advances in solar technology. In order to provide affordable electricity to low-income households, the government of Rwanda has pledged to achieve 48% of its overal electrification goals from off-grid solar systems by 2024. In this paper, we develop a cost-effective power generation model for a solar PV system to power households in rural areas in Rwanda at a reduced cost. A performance comparison between a single household and a microgrid PV system is conducted by developing efficient and low-cost off-grid PV systems. The battery model for these two systems is 1.6 kWh daily load with 0.30 kW peak load for a single household and 193.05 kWh/day with 20.64 kW peak load for an off-grid PV microgrid. The hybrid optimization model for electric renewable (HOMER) software is used to determine the system size and its life cycle cost including the levelized cost of energy (LCOE) and net present cost (NPC) for each of these power generation models. The analysis shows that the optimal system’s NPC, LCOE, electricity production, and operating cost are estimated to 1,166,898.0 USD, 1.28 (USD/kWh), 221, and 715.0 (kWh per year, 37,965.91 (USD per year), respectively, for microgrid and 9284.4(USD), 1.23 (USD/kWh), and 2426.0 (kWh per year, 428.08 (USD per year), respectively, for a single household (standalone). The LCOE of a standalone PV system of an independent household was found to be cost-effective compared with a microgrid PV system that supplies electricity to a rural community in Rwanda.

Since the first monkeypox (MPX) case was reported in humans in 1970, there have been several outbreaks of the disease. MPX is endemic in central and western Africa. MPX virus infection is confirmed using the conventional polymerase chain reaction, which detects the viral DNA in samples from the rash. Of concern is that the current outbreak has affected five regions of the world. Although MPX confirmatory tests are available worldwide, there are concerns about Africa's capacity to diagnose and contain the disease. The challenges faced by Africa include a lack of adequate laboratory infrastructure and health care workers, weak disease surveillance systems, and a lack of MPX knowledge among health care workers and communities. These challenges can be addressed by mobilizing resources for MPX virus testing, strengthening surveillance systems, collaboration among countries, training health care workers, task shifting, and engaging communities.

Social media messaging applications(i.e. WhatsApp, Facebook) have reached 2.3 billion users in 2019, with the majority of users emerging from developing countries. The high usage among emergent users opens the possibility of designing text-based interventions for social change but such interventions rely on experts (i.e. doctors, educators, and moderators) knowledge which is scarce in developing contexts. Expert knowledge can be scaled up using chatbots but more research is needed to support emergent users who need context-specific support such as local language interventions or may not have regular internet connectivity. Therefore to support the design of chatbot based interventions in low resource contexts, we built DIA a chatbot architecture for low resource contexts to scale expert knowledge and support localization. DIA is a human-chatbot (humbot) hybrid system that organically learns topic-specific knowledge and local language from user interactions. We built a preliminary version of DIA on WhatsApp and deployed it to mentor 38 teachers in a rural context of Côte d'Ivoire. Through our preliminary deployment, we show that DIA can help (1) build a data-set of a topic and language-specific dialogues (2) understand users' online smartphone usage through chat logs and (3) collect survey data for through conversational interaction.

It is evident that recently reported robot-assisted therapy systems for assessment of children with autism spectrum disorder (ASD) lack autonomous interaction abilities and require significant human resources. This paper proposes a sensing system that automatically extracts and fuses sensory features, such as body motion features, facial expressions, and gaze features, further assessing the children behaviors by mapping them to therapist-specified behavioral classes. Experimental results show that the developed system has a capability of interpreting characteristic data of children with ASD, thus has the potential to increase the autonomy of robots under the supervision of a therapist and enhance the quality of the digital description of children with ASD. The research outcomes pave the way to a feasible machine-assisted system for their behavior assessment.

The electrification of personal transportation holds great potential for lowering greenhouse gas emissions and reducing climate change. The promise of electric vehicles (EVs) to serve these goals has resulted in a broad range of supporting policies aimed at encouraging widespread EV adoption at both the state and federal levels in the United States and around the world. While the EV revolution and prospects of a world with ubiquitous EVs are impacting various industries and many aspects of daily life, strategic interactions between the power grid and EVs are crucial for a successful energy transition. However, managing the interplay between EVs and the power grid remains a challenge. Motivated by that tension, this paper surveys a variety of solutions, policies, and incentives that are focused on effectively managing EV charging behaviors. The paper’s objective is to explore these tools to ensure that EV owners have ultimate control over their personal vehicles while simultaneously allowing the power grid to mitigate adverse network impacts. Furthermore, this paper examines the role of charging infrastructure technology and its strategic placement in facilitating the seamless integration of EVs into the grid. Additionally, the paper highlights financial mechanisms associated with EV integration and discusses the consequences of these mechanisms.

Given the fierce competition that has come up because of evolving FinTech and e-payment industries in the global market, the credit card industry has become extremely competitive. To survive, financial institutions need to offer their credit card customers with more innovative financial services that provide a personalized customer experience beyond their banking needs. While we are witnessing this high competition that aims to provide better services to credit card holders, Africa risks remaining behind once again: in 2017, the World Bank reported that only 4.47% of Africans aged 15 and above hold a credit card. In this paper, we define and describe the steps that can be taken to build a behavioral-based segmentation model that differentiates African credit cardholders based on their purchases data. We focus on African customers and African financial institutions as (i) little has been done so far when it comes to understanding the spending behavior of African credit card holders; and (ii) because we believe that this segmentation will allow boosting credit card usage in Africa, thus allowing Africans to fully benefit from credit cards as other parts of the world do. The results of this research can help tailor the market campaign to make them customer- centric and reduce the associated marketing costs. We show the proposed approach at work using anonymized credit card data of one the leading banks in Egypt, the Commercial International Bank of Egypt.