The most recent advances in artificial intelligence, as concerns both software and hardware, are fostering a multitude of smart devices capable to recognize and to react to music, images, as well as to other “stimuli”. These autonomous things, from robots to cameras to healthcare devices, could exploit recent advances in Internet of Things and resort to pervasive and distributed computing techniques in order to avoid constant connection to the Cloud.
With artificial intelligence embedded in a great variety of communicating devices and machines, we are reaching the so called pervasive intelligence scenario, wherein machine and devices can communicate with each other independently of any human being. The proper integration of deep learning into these smart devices could boost definitively this trend into a common reality.
On one hand, local chips for deep learning may benefit Internet connectivity as well as proper and efficient pervasive and distributed computing techniques, in order to increase their local performance. This can be achieved by exploiting the well-known edge and fog computing paradigms, which do not suffer from the latency issues typical of traditional Cloud-based analyses. This is especially true because deep learning requires great computational power, which could be properly distributed and parallelized, and a great amount of data, which could also be available in the form of fast streams of data managed pervasively by a multitude of devices. On the other hand, deep learning techniques could help to improve the performance of both parallel and distributed computing techniques themselves, by finding out opportune strategies and mechanisms to efficiently distributed workload and tasks across different connected smart nodes.
The PerDL workshop aims to bring together practitioners and researchers working on pervasive computing and on deep learning, by soliciting contributions on, but not limiting to, the following topics:
- Advances in pervasive and distributed deep learning techniques and algorithms;
- Theories, models and novel algorithms for rendering deep learning suitable to pervasive and distributed computing;
- Novel applications of deep learning techniques in the context of pervasive and distributed computing;
- Technological innovations making possible the integration of deep learning and pervasive computing;
- Fog and edge computing techniques for deep learning;
- Researches to make the computational complexity of deep learning methods suitable for distributed devices;
- Studies to efficiently distribute and retrieve great amounts of data useful for deep learning algorithms;
- Deep learning techniques to improve the performance of pervasive and parallel computations.
The submitted papers may regard analytical, empirical, technological, or methodological themes, as well as a combination of these. The impact and influence of the contributions should be demonstrated in the context of both pervasive computing and deep learning. Papers applying known techniques from other fields are encouraged, provided that the main topics of the workshop are properly addressed.