Posted Content•
Predicting cell phone adoption metrics using satellite imagery.
TL;DR: This paper presents a method that uses publicly available satellite imagery to predict telecoms demand metrics, including cell phone adoption and spending on mobile services, and applies the method to Malawi and Ethiopia.
Abstract: Approximately half of the global population does not have access to the internet, even though digital connectivity can reduce poverty by revolutionizing economic development opportunities. Due to a lack of data, Mobile Network Operators and governments struggle to effectively determine if infrastructure investments are viable, especially in greenfield areas where demand is unknown. This leads to a lack of investment in network infrastructure, resulting in a phenomenon commonly referred to as the `digital divide`. In this paper we present a machine learning method that uses publicly available satellite imagery to predict telecoms demand metrics, including cell phone adoption and spending on mobile services, and apply the method to Malawi and Ethiopia. Our predictive machine learning approach consistently outperforms baseline models which use population density or nightlight luminosity, with an improvement in data variance prediction of at least 40%. The method is a starting point for developing more sophisticated predictive models of infrastructure demand using machine learning and publicly available satellite imagery. The evidence produced can help to better inform infrastructure investment and policy decisions.
Citations
More filters
01 Jan 2013
8 citations
TL;DR: In this paper, a technical and economic framework for the deployment of broadband networks in rural and remote areas is proposed, which takes into account the specific features of these regions and adapts services and network applications to meet the needs of these communities.
Abstract: Expanding broadband services represents a significant challenge for broadband operators, especially in light of the requirements related to the total cost of ownership of these technologies. In the last few years, this expansion has advanced significantly, but it still represents a challenge that must be overcome since there is a need to provide low-cost services to rural communities in remote areas. Issues related to geographical location, the low income of residents, and the lack of public infrastructural facilities lead to a disadvantageous relationship between the potential revenue for operators and the high costs of deploying infrastructure. Although there are several research endeavors in the literature aimed at addressing how connectivity can be provided, they do not discuss systems that take account of the specific features of these regions or that have adapted services and network applications to meet the needs of these communities. Thus, using dimensioning systems for the total cost of network ownership and taking into account capital and network operating expenses, this study establishes a technical and economic framework for the deployment of broadband networks in rural and remote areas. It also applies economic feasibility analysis techniques designed to assist decision making by interpreting the effects of any financial investment made and estimating the expected profits of the broadband operators. We also recommend the use of socioeconomic indicators to predict the potential social impact of this framework on the development of these regions. We employ a case study to demonstrate the operational features of the planned framework. Based on real data obtained from a municipality located in the Brazilian Amazon region, we show that it is possible to reduce the cost of subscribing to broadband services for end-users by reducing deployment costs and thus ensure that access to digital services can be equitably obtained.
7 citations
Posted Content•
TL;DR: In this article, the authors developed open-source software to test broadband universal service strategies which meet the 10 Mbps target being considered by the UN Broadband Commission and quantified the private and government costs of different infrastructure decisions.
Abstract: Internet access is essential for economic development and helping to deliver the Sustainable Development Goals, especially as even basic broadband can revolutionize available economic opportunities Yet, more than one billion people still live without internet access Governments must make strategic choices to connect these citizens, but currently have few independent, transparent and scientifically reproducible assessments to rely on This paper develops open-source software to test broadband universal service strategies which meet the 10 Mbps target being considered by the UN Broadband Commission The private and government costs of different infrastructure decisions are quantified in six East and West African countries (Cote D`Ivoire, Mali, Senegal, Kenya, Tanzania and Uganda) The results provide strong evidence that `leapfrogging` straight to 4G in unconnected areas is the least-cost option for providing broadband universal service, with savings between 13-51% over 3G The results also demonstrate how the extraction of spectrum and tax revenues in unviable markets provide no net benefit, as for every $1 taken in revenue, a $1 infrastructure subsidy is required from government to achieve broadband universal service Importantly, the use of a Shared Rural Network in unviable locations provides impressive cost savings (up to 78%), while retaining the benefits of dynamic infrastructure competition in viable urban and suburban areas This paper provides evidence to design national and international policies aimed at broadband universal service
2 citations
31 Mar 2023
TL;DR: In this article , the authors apply the ITU-R P.618-8 model with data from TRMM and Global Precipitation Mission (GPM) satellite to determine the level of attenuation and available link margin for a LEO system such as Telesat.
Abstract: In this paper, we apply the ITU-R P.618-8 model with data from the ITU-R P.837-7, Tropical Rain Measuring Mission (TRMM) and Global Precipitation Mission (GPM) satellite to determine the level of attenuation and available link margin for a LEO system such as Telesat. The specific and predicted attenuation for chosen six candidate ground stations (Abuja, Hartbeesthoek, Cairo, Longonot, Port Louis and Praia) is computed and results presented for 0.001%-1% unavailability time in a year. Setting a link margin of 0.36dB, the available link margin and the best candidate ground station for a LEO system such as Telesat is determined. The approach used can be implemented for other potential ground stations and LEO communication systems over Africa.
References
More filters
Proceedings Article•
01 Jan 2015TL;DR: In this paper, the authors investigated the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting and showed that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 layers.
Abstract: In this work we investigate the effect of the convolutional network depth on its accuracy in the large-scale image recognition setting. Our main contribution is a thorough evaluation of networks of increasing depth using an architecture with very small (3x3) convolution filters, which shows that a significant improvement on the prior-art configurations can be achieved by pushing the depth to 16-19 weight layers. These findings were the basis of our ImageNet Challenge 2014 submission, where our team secured the first and the second places in the localisation and classification tracks respectively. We also show that our representations generalise well to other datasets, where they achieve state-of-the-art results. We have made our two best-performing ConvNet models publicly available to facilitate further research on the use of deep visual representations in computer vision.
49,914 citations
TL;DR: The relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift are discussed.
Abstract: A major assumption in many machine learning and data mining algorithms is that the training and future data must be in the same feature space and have the same distribution. However, in many real-world applications, this assumption may not hold. For example, we sometimes have a classification task in one domain of interest, but we only have sufficient training data in another domain of interest, where the latter data may be in a different feature space or follow a different data distribution. In such cases, knowledge transfer, if done successfully, would greatly improve the performance of learning by avoiding much expensive data-labeling efforts. In recent years, transfer learning has emerged as a new learning framework to address this problem. This survey focuses on categorizing and reviewing the current progress on transfer learning for classification, regression, and clustering problems. In this survey, we discuss the relationship between transfer learning and other related machine learning techniques such as domain adaptation, multitask learning and sample selection bias, as well as covariate shift. We also explore some potential future issues in transfer learning research.
18,616 citations
"Predicting cell phone adoption metr..." refers background in this paper
...Several types of transfer learning have been surveyed in the literature (Pan and Yang, 2010), but inductive...
[...]
...Several types of transfer learning have been surveyed in the literature (Pan and Yang, 2010), but inductive transfer learning is a commonly applied approach, where the domain of two machine learning problems are the same, but the task is different....
[...]
Posted Content•
TL;DR: A statistical framework is developed that uses satellite data on lights growth to augment existing income growth measures, under the assumption that measurement error in using observed light as an indicator of income is uncorrelated with measurementerror in national income accounts.
Abstract: GDP growth is often measured poorly for countries and rarely measured at all for cities or subnational regions. We propose a readily available proxy: satellite data on lights at night. We develop a statistical framework that uses lights growth to augment existing income growth measures, under the assumption that measurement error in using observed light as an indicator of income is uncorrelated with measurement error in national income accounts. For countries with good national income accounts data, information on growth of lights is of marginal value in estimating the true growth rate of income, while for countries with the worst national income accounts, the optimal estimate of true income growth is a composite with roughly equal weights. Among poor-data countries, our new estimate of average annual growth differs by as much as 3 percentage points from official data. Lights data also allow for measurement of income growth in sub- and supranational regions. As an application, we examine growth in Sub Saharan African regions over the last 17 years. We find that real incomes in non-coastal areas have grown faster by 1/3 of an annual percentage point than coastal areas; non-malarial areas have grown faster than malarial ones by 1/3 to 2/3 annual percent points; and primate city regions have grown no faster than hinterland areas. Such applications point toward a research program in which "empirical growth" need no longer be synonymous with "national income accounts."
1,449 citations
"Predicting cell phone adoption metr..." refers background in this paper
...Many papers have focused on using nightlight luminosity data to assess questions relating to economics (Henderson et al., 2012, 2011), human development (Bruederle and Hodler, 2018), urban extent (Zhou et al., 2015), conservation (Mazor et al., 2013), atmospheric composition (Proville et al., 2017)…...
[...]
Proceedings Article•
01 Jan 2019TL;DR: To demonstrate adapter's effectiveness, the recently proposed BERT Transformer model is transferred to 26 diverse text classification tasks, including the GLUE benchmark, and adapter attain near state-of-the-art performance, whilst adding only a few parameters per task.
Abstract: Fine-tuning large pre-trained models is an effective transfer mechanism in NLP. However, in the presence of many downstream tasks, fine-tuning is parameter inefficient: an entire new model is required for every task. As an alternative, we propose transfer with adapter modules. Adapter modules yield a compact and extensible model; they add only a few trainable parameters per task, and new tasks can be added without revisiting previous ones. The parameters of the original network remain fixed, yielding a high degree of parameter sharing. To demonstrate adapter's effectiveness, we transfer the recently proposed BERT Transformer model to 26 diverse text classification tasks, including the GLUE benchmark. Adapters attain near state-of-the-art performance, whilst adding only a few parameters per task. On GLUE, we attain within 0.4% of the performance of full fine-tuning, adding only 3.6% parameters per task. By contrast, fine-tuning trains 100% of the parameters per task.
1,308 citations
TL;DR: In this paper, satellite data on lights at night is used to augment existing income growth measures, under the assumption that measurement errors in using observed light as an indicator of income is uncorrelated with measurement error in national income accounts.
Abstract: GDP growth is often measured poorly for countries and rarely measured at all for cities or subnational regions. We propose a readily available proxy: satellite data on lights at night. We develop a statistical framework that uses lights growth to augment existing income growth measures, under the assumption that measurement error in using observed light as an indicator of income is uncorrelated with measurement error in national income accounts. For countries with good national income accounts data, information on growth of lights is of marginal value in estimating the true growth rate of income, while for countries with the worst national income accounts, the optimal estimate of true income growth is a composite with roughly equal weights. Among poor-data countries, our new estimate of average annual growth differs by as much as 3 percentage points from official data. Lights data also allow for measurement of income growth in sub- and supranational regions. As an application, we examine growth in Sub Saharan African regions over the last 17 years. We find that real incomes in non-coastal areas have grown faster by 1/3 of an annual percentage point than coastal areas; non-malarial areas have grown faster than malarial ones by 1/3 to 2/3 annual percent points; and primate city regions have grown no faster than hinterland areas. Such applications point toward a research program in which "empirical growth" need no longer be synonymous with "national income accounts."
1,216 citations