Our core course Machine Learning for Pervasive Computing (ELEC-E7260) focuses on advanced Machine Learning approaches prominently applied in Pervasive Computing domains.


  • Food Object Detection and Recognition with Region Proposal Networks (MSc)
    Janaki Koirala, ongoing
  • Edge-based recommender system for city visitors exploiting egocentric images and video (BSc)
    Andreas Hitz, ongoing
    Hitz This thesis aims to design and conceptualize a recommender system for city visitors, which is based on egocentric vision, i.e. images and videos obtained from head or body mounted cameras. From the images, videos and associated information, e.g. time, location or textual content description such as activity or mood, the system extracts information to generate personal recommendations for city visitors. Therefore a conceptional architecture is designed and described, covering also technical details, e.g. communication technologies, processing tools and storage solutions. Other discussed aspects are how to guarantee privacy in such a data-intensive architecture, and algorithmic considerations - in particular, how to distribute processing load using a cloud- or edge-based infrastructure. In addition, a comprehensive literature review on related work is provided.
  • Walking Speed detection from 5G Prototype System (MSc)
    Bahareh Gholampooryazdi, 05.2017
    Bahareh While most RF-sensing approaches proposed in the literature rely on short-distance indoor point-to-point instrumentations, actual large-scale installation of RF sensing suggests the use of ubiquitously available cellular systems. In particular, the 5th generation of the wireless communication standard (5G) is envisioned as a universal communication means also for Internet of Things devices. This thesis presents an investigation of device-free environmental perception capa- bilities in a 5G prototype system in two cases; walking speed and human presence detection, and elaborate a comparison with the former case and acceleration sensing analysis. This thesis attempts to analyze the perception capabilities of 5G system in order to recognize human mostly common activities and presence detection near transceiver devices which the instrumentation exploits a device-free system capable of detect activities without carrying devices capitalizing on environmental RF-noise. This is done via the study of existing and related literature. After that, the imple- mentation and evaluation of walking speed and presence detection is described in details. In addition, evaluation consists of utilizing a prototypical 5G system with 52 OFDM carriers over 12.48 MHz bandwidth at 3.45 GHz, which we consider the impact of the number and choice of channels and compare the recognition performance with acceleration-based sensing. It was concluded that in realistic settings with five subjects, accurate recognition of activities and environmental situations can be a reliable implicit service of future 5G installations.
  • Transfer learning in emotion recognition (BSc)
    Niklas Strengell, 05.2017
    Strengell The purpose of this thesis is to review and discuss automated emotion recognition and transfer learning. Firstly, emotional theories are discussed and the various modalities from which emotions can be recognized are introduced. Secondly, transfer learning, which is a machine learning technique for transferring previously learned knowledge, is discussed and explained in detail. Lastly, new research which exploits transfer learning techniques in automated emotion recognition is reviewed and the future of research is discussed.
  • Blockchain as a distributed database for edge-supported IoT (BSc)
    Juuso Mikkonen, ongoing
    Mikkonen Internet of Things (IoT) is an emerging trend in many industrial and consumer applications. IoT inherently poses new challenges to technical implementations due to its distributed nature and large amounts of gathered data. A fundamental challenge that must be addressed in IoT applications is the storage of data gathered by edge devices in the network. The devices themselves are typically not very performant and have little capacity to store data. Industrial IoT device networks of large amounts of edge devices can extract new data at a very high rate. The data can be a key element of the business logic of the IoT network. A distributed database implemented utilizing blockchain technology is one option for IoT data storage. Decentralized nature and inherent security are traits that make the technology an appealing choice for distributed storage. Blockchain technology also poses some challenges not faced by more traditional distributed databases. Current implementations like the Bitcoin protocol are severely limited by their performance. Most often there exists a trade-off situation between throughput and latency. Recent development in the field of blockchain technology has brought many possible improvements to the blockchain model. The utilization of new protocols could potentially solve many of the scalability issues and trade-offs inherent to previous technologies. This would in turn enable the use of blockchain in many large-scale industrial applications. A successful solution to IoT edge storage would have a big effect on industrial IoT solutions. It would enable the use of cheaper and more simple devices while simultaneously extending the size of device networks. The security of the solutions could also be improved in the process enabling the use of IoT in fields with high privacy requirements.
  • Application of Internet of Things Framework On Enhanced Lean Construction Management (MSc)
    Dinesh Hyaunmikha, ongoing
    Mikkonen Internet of things is an ever emerging field with its application growing in all areas. Smart housing and smart cities are some of them with intelligent products and connected devices through internet. Currently, research is extending towards application of IoT communication framework for automation of production control in construction industry as well. Information flow and synchronization between site teams, managers, various contractors and suppliers still remains a major factor of delay in construction workflow. Construction management systems, such as “VisiLean” tries to overcome some of these overheads by providing a visual model (BIM – “Building Information Model”) along with the management workflow. This main aim of this thesis is to study the practical applications of IoT communication framework on construction management systems on scenarios such as automated procurement, tracking resources or production planning and control. Such information would then be represented in the relevant part in the BIM automatically.

Maarintie 8, 00076 Espoo; +358 (0)50 4666941; ambience-elec@aalto.fi