New! EPSRC Subawarded Smartcity project (Jan 2020 - present): Read more details here.

Machine learning in IoT - TinyML/Sensys-ML/UltraML

Anomalies detection in distributed networks and cyber-physical systems (CPS) is a challenging task and require multi-objective optimisation. In recent years, Federated machine learning is used for privacy-preserved machine learning. In Federated Learning many decentralised distributed computing nodes (edge or servers) train a local model on local data and then local model are combined centrally to generate a global model. There are different frameworks and algorithms to implement a Federated Learning system. In this project, the aim to design and implement an efficient Federated Machine Learning Algorithm for anomaly detection and prediction in IoT.

Recent Publications: 

A. Feraudo, P. Yadav, V. Safronov, D. A. Popescu, R. Mortier, S. Wang, P. Bellavista, and J. Crowcroft:

Colearn: Enabling federated learning in mud compliant IoT edge networks. The 3rd International Workshop on Edge Systems, Analytics and Networking (EdgeSys20). New York: ACM, April 2020 [PDF]

Colby R. BanburyVijay Janapa ReddiMax LamWilliam FuAmin FazelJeremy HollemanXinyuan HuangRobert HurtadoDavid KanterAnton LokhmotovDavid A. PattersonDanilo PauJae-sun SeoJeff SierackiUrmish ThakkerMarian Verhelst, Poonam Yadav:

Benchmarking TinyML Systems: Challenges and Direction.  MLSys 2020 (CoRR abs/2003.04821)(2020)

Anomaly detection, Device Fingerprinting and Traffic Filtering in IoT

Internet-of-things (IoT) enabled smart build environments are an accurate representation of complex and dynamical systems. The diversity and heterogeneity of components in the IoT, not only make the ecosystem extremely difficult to analyse and validate but also make it hard to build both secure and accountable. To address these issues, the Internet Engineering Task Force (IETF) has taken the initiative to bring a standard (RFC8520), which will encourage manufacturers of IoT devices to provide a Manufacturer Usage Description (MUD) for their IoT devices. In our work, we explore a deployment scenario of MUDs in domestic settings and proposing an MLogger application which runs on a local router. MLogger enforces user-defined traffic filtering policies along with the MUD policies for each IoT devices. Our solution not only aims to provide a better fine-grained traffic filtering locally but also enables a user-defined control and accountability at the edge of the network.

Recent Publications: 

[New] Poonam Yadav, Angelo Feraudo, Budi Arief, Siamak F. Shahandashti Vassilios G. Vassilakis:

Position paper: a systematic framework for categorising IoT device fingerprinting mechanisms. ACM AIChallengeIoT 2020 [PDF] [bibtex]

Poonam YadavQi LiRichard MortierAnthony Brown:
Network service dependencies in commodity internet-of-things devices. ACM IoTDI 2019: 202-212 [PDF] [bibtex]

John MooreAndrés Arcia-Moret, Poonam YadavRichard MortierAnthony BrownDerek McAuleyAndy CrabtreeChris GreenhalghHamed HaddadiYousef Amar:
Zest: REST over ZeroMQPerCom Workshops 2019: 1015-1019 [PDF] [bibtex]

Poonam YadavVadim SafronovRichard Mortier:
Enforcing accountability in Smart built-in IoT environment using MUDBuildSys@SenSys 2019: 368-369

Diana Andreea PopescuVadim SafronovPoonam YadavRoman KolcunAnna Maria MandalariHamed HaddadiDerek McAuleyRichard Mortier:
"Sensing" the IoT network: Ethical capture of domestic IoT network traffic: poster abstract. SenSys 2019: 406-407

Recent Projects

EPSRC Vacation Internships (Jun 2020 - September 2020): Find details here.

EPSRC DataBox @ Cambridge University (March, 2017 - May 2020)

In this project we explored the development of the Databox as means of enhancing accountability and giving individuals control over the use of their personal data.

The Databox envisions an open-source personal networked device, augmented by cloud-hosted services, that collates, curates, and mediates access to an individual’s personal data by verified and audited third party applications and services. The Databox will form the heart of an individual’s personal data processing ecosystem, providing a platform for managing secure access to data and enabling authorised third parties to provide the owner with authenticated services, including services that may be accessed while roaming outside the home environment.

Collaborators: PI -  Dr. Hamed Haddadi (Imperial College London), Dr. Richard Mortier (University of Cambridge) and Professors Derek McAuley, Tom Rodden, Chris Greenhalgh, and Andy Crabtree (University of Nottingham).

Follow updates on the project here.

UbiquityLab @ Cambridge University (Since Oct, 2015 - March 2017)

UbiquityLab is a novel platform for creating and executing interactive experiments on the Internet. More broadly, the platform aims to promote online experiments for behavioral research in the social sciences by simplifying their creation and execution. The current version focuses on game-theoretic experiments.

In general, the platform accommodates two types of interaction. The first type is real-time interaction: subjects continuously act and respond to actions in a matter of seconds, and the whole experiment usually lasts no more than a few hours. For example, an experiment utilizes this type of interaction as subjects choose efforts and receive the results of their choices within 20 seconds over 40 consecutive rounds. The second type is extended interaction: subjects engage with each other irregularly and often in sequence, and the whole experiment may last days, weeks or even months. The current version emphasizes the support for real-time interactive experiments.

EU FP7 Citizen Cyberlab @ Imperial College London (July 2013 - September 2015)

Citizen Cyberlab: The central focus of the research project is understanding creativity and learning in on-line citizen science. To explore these aspects of citizen science, we are evaluating existing on-line collaborative environments and software tools to assess their role in supporting and stimulating creative learning, as well as examining the best practices of current Citizen Science projects.

TSB OpenShare@Imperial College London (July 2013 - April 2014)

The OpenShare platform is built to allow a group of collaborating entities to interact with media data for processing complex shared workflows. It will support provisioning of computational capacity to undertake these workflows on cloud computing infrastructure. OpenShare also provides necessary tools and services to control access to the data. It also ensures that both data and the tasks to be carried out on it are only available to authorised individuals or groups.

Diversity and Equality @ ACM-W UK(Since Nov, 2014)

Recent years have seen an increase in diversity initiatives worldwide with different organisations emphasizing the need for a 50-50 male and female workforce distribution. Different initiatives have been proposed to bring women on boards, especially in STEM (Science, Technology, Engineering, Mathematics) and make them comfortable in the current working environments. To understand the impact of these initiatives, ACM-W UK conducted an online survey. This work presents the useful insights drawn from the results of the survey and also our recommendations for STEM and computing fields to increase female numbers in their programs.

Previous Projects

  • NERC FUSE @ Imperial College London (Jan 2012 - June 2013)
  • Emergent Synchronisation@Imperial College London (July 2010 - Oct 2012)
  • Distributed Database@IBM  Research (Sept 2011 - March 2012)
  • Power Efficient Networking Stack  for WSN@Imperial College London (July 2009 - July 2010)
  • Adaptive Routing WSN@Imperial College London (Oct 2007 - June 2009)
  • Face Predict@IIIT, Allahabad (July 2006 - Jan 2007)
  • FlexiMote@IIIT, Allahabad (July 2006 - Sept 2007)