Machine-to-Machine Learning based framework for ad-hoc IOT ecosystems

We put forward a proposal for a new kind of framework for IOT networks, which can help to quickly setup, build and deploy dedicated IOT networks (we called them "ecosystems") for research or surveillance purposes.The proposal includes three components: the software which implements the framework itself, the machine-to-machine learning (M2M) component of the network, and the customization choice of opting either for a centralized compute & storage workstation or a block-chain powered decentralized server. The paper emphasizes on the possible implementation of a centralized IOT "ecosystem" and leaves the decentralized version for future discussion. The network is adhoc in nature to allow for devices for an easy "connect-&-transmit" process. Given a centralized compute-storage server, a user setups multiple devices like cell phones, camera-microphone enabled smartphones, high quality video recorders, and registers the specific devices to the central server which we call "Control". Once these devices are registered, the Control deploys the built M2M model to learn about the data stream as generated by the registered and now live endpoints of the ecosystem built.The M2M model can be more robust alternative for typical anomaly detection models. In addition to this, the data stream can be studied and analysed for specific patterns during specific periods corresponding to exceptional situations like device malfunctions. We introduce a concept of a trust factor for each of the live devices in ecosystem. For a device to have a low degree of trust, the model will be sensitive to anomalies in data stream from that device. For devices with high trust, the model will assume a stance of variable doubt based on its data stream.