from 18:45 pm to 19:30 pm
The existing InternetofThings ecosystem already provides offtheshelf tools to create smart devices, integrate them with a variety of sensors, and connect them through Internet using efficient networks. The emerging challenge is actually what to do with the data that we are reading from those sensors. Do we have to store all of it? Only for a small period of time? What type of knowledge can we extract from it? How can we leverage that knowledge?
The current approach to solve storage problems rely on using any of the existing IoT platforms. Most common platforms provide basic functionality to record data from sensors, access the raw data streams, and sometimes even provide basic transformation logic for collected events.
These functionalities allow users to monitor sensor data (streams) and perform basic univariate analysis on the existing streams. However, more complex analytics usually requires an offline batch processing approach, (and significant efforts of domain experts and machine learning specialists) the help of a domain expert, and a machine learning specialist. Moreover, the storage requirements grow as platforms become longtermarchives for our data. This situation imposes a huge step between the users that would like to extract more value from its data and the infrastructure required to perform such analysis.
To aggravate even more the situation, the world of IoT has its own inner complexity related to the method used to transmit data and the quality of the data being transmitted. That translate into assuming that the events are likely to arrive outoforder, some measures may not appear all times due to sensor failures, outofsync events may appear when a device reconnects to the network, etc. Given this situation, we decided to approach the problem from another point of view and build a streaming infrastructure that offers unsupervised machine learning for both univariate and multivariable analytics.
In this talk, we present the Novelti architecture, a streaming architecture for intelligent IoT analytics that reduces the complexity of the required infrastructure and simplifies the interaction between the users and the underlying value found in their streams. The talk will cover first the singularities of working on an IoT oriented environment. Afterwards we will describe our approach to streaming machine learning, and finally we will present the main components of our architecture and how the user is able easily extract knowledge and value from its streams. If you are building your own IoT device, or designing the next wearable, or managing a complex IoT infrastructure, this talk will enlighten some interesting ideas.