"Glocal" and continuous Machine Learning for a society of intelligent devices

We are immersed in a society of devices (mobile phones, tablets, wearables, etc.) that are transforming us at an incredible speed, changing the way we live or even how we interact with each other. The progressive incorporation of new sensors on these devices and their connection to networks allows them to access information both referring to their local environment and to the whole world. This opens important opportunities regarding the development of systems and applications that use this information for our benefit in an increasing number of domains: health, leisure, education, sport, social interaction... However, it also makes more necessary the use of machine learning strategies to face the important challenges which are specific to this society of devices: huge amount of information, limited processing, hardware disparity, noise on data which is inherent to the unlimited boundary conditions in which they are captured, singularities in the behavior of the user, privacy, etc.

In this context, the concept of "glocal" learning: i.e. learning and local adaptation of models, in the device itself, which can be further improved at the global level, in the cloud, bringing together which has been learned locally. Now, if there is something that characterizes this society of devices is its high heterogeneity and dynamism, both in terms of users and the hardware itself. For this reason, to the previous "glocal" learning we must add the need to be carried out continuously, through a cyclical process of global consensus and local adaptation that can be repeated indefinitely over time.

The formulation of this continuous glocal learning is the main objective of this project. Finally, the fact that between the local and the global level only models, and not data, are moved allows overcoming difficulties related to the transfer and privacy of the information that is being moved between the devices and the cloud. The emergence of continuous glocal learning would allow us to begin to build a set of strategies, which would also include Deep Learning, which would make possible to solve complex problems effectively, and with which the world of applications based on sensors might be revolutionized.


The specific objectives of the Project are the following:

  1. To define a strategy for glocal learning, inspired in consensus algorithms, combining local and global information in order to reach an asymptotic consensus between models stored in neighborhood devices.
  2. To define a strategy for continuous learning which will allow the temporal evolution of the models stored in the devices.
  3. With the aim of showing the results obtained in the framework of local and continuous learning in devices, we will develop three technological showcases. This technologies will be focused on: (a) biometric identification of faces without restrictions, (b) identification based on walking way, (c) construction of radio maps for indoor wifi localization.
  4. Dissemination of the results by means of publications in high impact journals, communications in high prestige conferences and technological events in cooperation with the industry sector.