IoT where value increasingly comes from information

IoT is a pervasive connectivity of things with internet capability. IoT was originated in 1999 by kelvin Ashton. The ‘things’ can be data, hardware, software, service and mixture of these. According to Gartner’s hype cycle prediction (for emerging technology) IoT is at peak position. A very large number of ‘things’ will join the IoT due to various tipping points like widespread wireless connectivity,low power and cost sensors technology, large cloud based storage and the large number of IP address with IPv6. The figure for the expected number of IoT devices in 2020 from the Cisco EMC & IDC and Gartner are shown in fig.1. Cisco also figures out that 50 billion is just 2.77 percentage of potentially connectable things that are estimated approximately 1.8 trillions. In a global economy and society where value increasingly comes from information knowledge and services the data collected from these huge number of devices represents the most abundant, valuable and complex raw material in the world.Just connecting ‘things’ and collection of data doesn’t fulfill true promises of IoT. Analysis of these data and information with great analytic power can provide real values but in the conventional programmable computing approach, the data is guided through a sequence of prearranged process to provide the results. This rigid process limits the IoT in acknowledging many problem of complex, real time, evolving world. The information extracted with these devices are complex, high volume and unpredictable,so analysis of these information with prearranged set up never suits for efficient results.This limitation can be explained with example of one application scenario of weather prediction. There are number of usage to weather predictions. Weather warnings are important forecasts because they are used to protect life and property.  Application scenario: Let temperature, humidity, wind speed, visibility, rainfall, pressure and wind direction sensors are deployed on number of stations and after analyzing the data collected from sensors prediction of rain, fog, drizzle, snow and thunder is done.  Processing these data with programmable computing will never give the needed accuracy in prediction as we are underlying the facts of aging effects, environmental condition effect on deployed sensors. In most computational algorithm and conventional learning techniques it was considered that these data generated from fixed probability distribution and may called stationary, but due to these effects data from IoT devices are non stationary. The stationarity means here is that the data are generated from a fixed, even though not known probability density function. But in case of practical applications, the data generated from sensors this type of assumption is not valid, because the process generating these types of data may no longer stationary. The probability distribution function (Pdf) of data is evolving or drifting with the time or we can say that non-stationary data is generated.For enabling any IoT application sensors are key components and are one of the major hardware component needed to form IoT systems. Thinking about the intelligent cyber physical systems, for environmental monitoring and forecasting, predicting energy demands, and recommendation system etc. always needs processing or analyzing the non-stationary data generated from the network of sensors. Sensors collect the data from the physical world and data is transmitted to a control location for processing. There are number of application fields that closely related to data collected from sensors in IoT domain shown in fig.2.  and various application related to the above fields are shown in the fig. 3 for making world smart. For developing IoT system for these application various sensors are deployed for data collection. Sensors may deployed at remote location, in the home, devices, on body depending upon the application. Let for detection of forest fire IoT systems the temperature, humidity sensor and gas sensor for carbon monoxide (CO) and Carbon Dioxide (CO2) are deployed at various station of forest, so sensing unit is the primary concern for any IoT system.Generally sensors are deployed in remote, harsh environments for significant period of time, so it is not possible to carry out maintenance on the sensor nodes after installation. Non-stationary property in the sensing unit is present due to the aging effects, faults in the sensor. Apart from it environment under monitoring also changes with time due to climate or whether change and loses sensitivity with time.In discussed application scenario of weather forecasting in section I, IoT system is dealing with basically classification problem that are utilized in forecasting decision of rain or no rain. So it may happen that it can provide wrong classification due to underlying the fact of non-stationarity of the data, and the subsequent wrong decisions. The accuracy of applications discussed in fig.3 or many countless application with IoT may effected with ignoring non stationarity or may provide wrong decisions for forecasting and false predictions.  Novelty detection,fraud detection, anomaly detection in industrial process monitoring, real time monitoring and control of some automated process and fault detection in techniques are directly effected by non stationary phenomena.It is very necessary to adapt the change in the data stream coming from sensor nodes in IoT systems applicable for monitoring and forecasting, predicting to provide best results. For achieving this, the primary requirement is that the learning model for IoT data should be intelligent enough to sense these changes from data itself. Learning in IoT domain process is shown in fig. 5. IoT data collected from nodes are stored through gateways and preprocessing is done.Now the process of learning with data starts, data can provide much more information by classification, clustering, pattern recognition, regression and many more depending upon the application. Knowledge base is prepared from result of applied algorithms and analysis is done for taking further decisions, and actuation based on decisions. Intelligence to the model can be provided with the three layers shown in fig. 5, in these layers the learning of the model is done with data itself.IoT systems merely incomplete without intelligence because all application environment for IoT systems are non stationary and learning in non stationary environment requires prominent intelligence so that model can learn with data itself and adapt with these changes. In fig. 5 learning in IoT domain in general is compared with the application scenario discussed in section I, here IoT sensor date is collected and features based on the data extracted. System is about prediction so the classification algorithm is used for classification of features and depending upon this the predictions is done. It is very much clear that the learning model is key part for IoT system intelligence. % FigureFor increasing the accuracy of such types of application based systems it is highly necessary for learning the data accurately the algorithm used should operate on transient stream pose several additional challenges such as the necessity to process instance by instance, availability of limited computational resources, memory and time and prediction being near real time or even real time. In this part the process of intelligent learning in IoT domain is discussed and will present the algorithmic and application based literature for non stationary learning in IoT domain in upcoming section. We describe the two primary families of strategies generally used for learning in non stationary environment. These two families are generally referred to as active and passive approaches. We also present and review algorithms for various problems like classification and regression. Finally, we describe the open problems for current and future research for learning in non-stationary environments for Iot domain, benchmark data-set and tools.