Ganes Kesari

Co-founder and Chief Decision Scientist

Ganes Kesari is a co-founder and head of analytics at Gramener, a data science company. Gramener solves business problems for its 100+ clients by identifying data insights and presenting them as data stories. Ganes advises executives at leading enterprises and NGOs on data science. He helps transform organizations through advisory in building teams and adopting a culture of data. He is a TEDx speaker, global corporate trainer and author. He is on a mission to simplify data science and help everyone understand its true potential.

WATCH LIVE: November 2nd at 11:00 am

Ganes Kesari

BRIEF: Today, no easy avenues exist to understand demographic or socio-economic insights in most countries. National census and sample surveys fail to capture the latest ground realities. Satellite imagery can be a powerful aid when viewed through the lens of deep learning. Geospatial analytics enabled by computer vision can help answer important questions, and save human lives.

DESCRIPTION: In most countries around the world, it is tough to base policy and socio economic decisions on data. Large governmental initiatives such as national census are collected once every 10 years. Despite the best of intentions, they aren’t comprehensive and quickly get out of sync with reality. Satellite imagery offers an alternate ground truth, that is accurate at highresolution, available across periods as a time series, easily accessible and is relatively economical. While this is a rich source of visual information, the challenge has been in processing images and generating useful insights. The advances in deep learning have helped solve this last hurdle, placing enormous power in our hands for socio-economic data analytics. Our work is inspired by Stefano Ermon et al, who used night light as a proxy to detect poverty in Africa (http://sustain.stanford.edu/predicting-poverty). Freely available high-resolution satellite imagery was combined with ground-truth survey data and labelled data from sources like Open Street Maps. By extracting the spatial attributes, a Deep Learning architecture was used to identify useful features from the maps such as buildings, tree cover, water bodies, and population density. This was used to arrive at important insights that could drive policy and socio-economic decisions. A real-world implementation of this work will be presented with a live demo of the results. The promising areas of application will be discussed to illustrate the potential to save lives. The target audience for this talk will be machine learning beginners, business leaders and data science practitioners. The talk outline will present the problem statement, solution approach and results in a way that is accessible to this audience. The deep learning architecture overview and modelling considerations would be shared to highlight the ML applications, but without getting too technical. The audience will understand what problems can be solved, how they should approach it, and where they can apply it.