Internet of Things (IoT)


Internet of Things (IoT)

Our environment is becoming permeated by an enormous amount of devices that generate data about our reality. Recent predictions vary in that between 20 and 200 billions of such devices will be connected to the Internet by 2020 as a part of the so-called Internet of Things (IoT). IoT has the ambition to interconnect smart devices across cities, industries, vehicles, appliances, retail, healthcare and other domains. In the agri-food sector only, it is estimated that by 2035 more than a billion of in-situ devices (sensors, cameras, etc.) will be installed across agricultural fields worldwide, generating vast amounts of data. Thus, a major challenge for data scientists and computer scientists of today and in the future is to extract knowledge buried in big data, i.e. to perform big data analytics.

Big Data Analytics

Can we make ultra-efficient and environmental-friendly future farms on account of huge amounts of data being collected (e.g. from in-situ sensors and cameras, environmental and climate data, etc.)? Can farmers take advantage of densely deployed in-situ sensor networks to sow seeds at variable spatial densities across a field – at very fine spatial resolutions – in order to “match” the complex variable nature of soil underneath, and thus take the best out of each piece of land? Can they use the in-situ sensors’ data to accurately predict pests and disease occurrence, and accordingly make efficient and eco-friendly, spatially variable spraying of pesticides? Can farmers take into account climate change data and weather forecasts data to make optimal decisions as to which crops to plant and at which day in a given season? Can they accurately predict prices for each fruit, vegetable, and cereal type at the market to make an optimal plan for each season or month? Can we make a platform which allows farmers to help each other by sharing their experiences and practices to their mutual benefits, while their private and sensitive data is not compromised? Finally, can we really do all this while constantly taking care of our planet, by maintaining and improving biodiversity and preventing or slowing down global warming and other negative climate changes?

 

The data needed to achieve this will certainly be there, and in gargantuan amounts. Indeed, it is estimated that, by 2035, more than a billion in-situ devices (sensors, cameras, etc.) will be installed across agricultural fields worldwide, while huge amounts of data about weather and climate are being collected on a daily basis. However, the main challenge then is how to extract knowledge from this big data in a time-efficient manner, such that farmers can effectively use it to make better decisions (e.g. decisions on what/when to plant, how dense, when to spray, etc.) The BioSense Institute develops mathematical and engineering tools to respond to this challenge. These tools go far beyond conventional ICT systems for data acquisition, storage and processing, and they rely on advanced methods in distributed optimization, machine learning, and network science. Specifically, our contributions include efficient distributed gradient optimization methods which converge with significantly less resources spent (inter-computer communication cost and computational cost) with respect to previously proposed distributed gradient methods, where savings go up to an order of magnitude, as well as optimal design of distributed inference algorithms through large deviations analysis. In terms of applications, we have developed machine learning based solutions for several real world agricultural use cases, including crop yields prediction based on environmental readings, soybean varieties portfolio optimization, and classification of small agricultural fields using satellite imagery.

image001CONDENSE concept: Can we communicate “big knowledge” instead of big data?

The vision of the future universal, integrated agri-food ICT platform assumes that huge amounts of the Internet of Things (IoT)-generated data (in-situ sensors, cameras, etc.) is transmitted through the communications infrastructure (cellular networks, optical networks) to the Cloud; only after that, these data are processed to generate actionable knowledge (the learning process). But what is behind this obscure concept of “Cloud”? Can it really accommodate ever-increasing amounts of data which are being generated every day? Even more importantly, can our communications infrastructure really communicate all these data to the Cloud? Regarding the former question, behind the concept of Cloud, in real world we actually have a vast number of data centers which consume significant amounts of energy to store, but also process the data (on the order of 3 percents of the globally consumed energy!) Regarding the latter issue of the communications infrastructure, it is essentially a collection of wireless and optical links through which we can transfer only finite amounts of information bits. Therefore, a natural question which arises is: Can we really go on forever with transferring these vast amounts of raw data through the communications infrastructure to the cloud, and, only after that, seek to extract knowledge from it? (Fig. 1-a).

 

We argue that, at some point in the future, such a solution may eventually become unsustainable. To cope with this challenge, BioSense – jointly with its partner institutions (Universities of Oxford, Tartu, Balearic Islands, Aalto, and the Technical University of Istanbul) – proposes the CONDENSE concept. CONDENSE brings a radically new idea to involve the communication infrastructure in knowledge acquisition. Instead of merely transferring data from the IoE sources to the cloud for knowledge extraction (learning), the communication infrastructure actively participates in learning: data traversing from the IoE sources through photonics and wireless links is processed enroute, by carefully designed modules in hardware and software (Fig. 1-b). This direct extraction of knowledge, as opposed to just transmitting data, will lead to a dramatic reduction in communications and storage costs, providing a sustainable solution for the future. Once realized, CONDENSE technology will dramatically change the landscape of telco/IP and big data ecosystems, boosting data analytics and IoE markets in the years to come. CONDENSE requires a highly interdisciplinary effort, involving machine learning, applied photonics, and wireless communications.

Fig. 1-a)
Fig. 1-b)

Fig. 1. The conventional IoT/Cloud integration can be represented by three major blocks (Fig. 1-a). The data layer contains billions of sensing and other data-generating devices. The communication layer provides mere data transfer to the data layer by essentially uploading the generated data to the learning layer (Cloud) in its entirety. The learning layer contains data centres which provide storage and processing capabilities, enabling learning algorithms to extract knowledge from big data. As opposed to this conventional solution, CONDENSE concept (Fig. 1-b) proposes a generic and flexible mechanism to extract knowledge within the process of communicating data, i.e., to harness the communication infrastructure to retrieve knowledge from the IoT devices rather than simply communicating all collected data.