Screenshot 2019-06-19 19.10.16.png

Professor Weisi Guo 
Chair Professor of Human Machine Intelligence

 Director of Smart Living Grand Challenge
Head of Human Machine Intelligence Group
Visiting Fellow at the Alan Turing Institute
​Cambridge: MEng, MA, PhD 


Office: F411 (1st floor iMech, B83)
Lab: F104 (1st floor DARTeC B105)

Centre for Autonomous and Cyber-Physical Systems, SATM, 
Cranfield University, 


Tel: 01234 758304


Webpages: Cranfield, Turing

Google Scholar, UKRI


  • Twitter
  • LinkedIn

Vision & Expertise - As human society is becoming increasingly interconnected, human and machine intelligence is more closely interfaced than ever. My expertise is in connected intelligence systems, comprised of: machine learning, communication networks, and socio-technical assets. In particular, I research future AI and information networks for challenging environments or tasks, whilst integrating human domain knowledge and providing trustworthy safety assurance. The research areas are:

  • Cyber - Networked Autonomy in Extreme Environments - We see increasing need and enabling technologies for distributed autonomy, but these are increasingly used in extreme or adversarial environments. Our research focus on developing secure communications to enable secure autonomy. We do so at the physical communication level using 2 primary mechanisms:
    1) Physical & Graph Layer Security - physically modifying and leveraging on real-world physics to generate digital cipher keys (see our EPSRC funded world first in Graph Layer Security)
    2) Molecular Signal Encoding - biochemically encoding digital signals to achieve molecular communications as a new transport medium for future 6G+ communications (see our Royal Society & USAF research in Molecular 6G

    Our research in networked autonomy then leverages distributed capabilities to investigate:
    3) Trustworthy Autonomy - establish human trust with machine intelligence through explainable deep learning, where we focus on meta-symbolic models and physics-informed models with brain inspired architectures for interpretation (see our EPSRC funded research in explainable AI for 6G and trustworthy AI for mass autonomy)
    4) Green Federated Learning - develop compression techniques to reduce energy consumption of neural networks using federated real-time compression (see our Nature Computation Science work on Random Sketch Federated Learning

    Key Grants: [UKRI-EPSRC] TAS-S Trustworthy Autonomous Systems-Security (EP/V026682/1, 2020-24, Co-PI)
    [EPSRC] Mobility as a service: MAnaging Cybersecurity Risks across Consumers, Organisations and Sectors (EP/V039164/1, 2021-23, CI)

    [H2020] DAWN4IoE - Data Aware Wireless Network for Internet-of-Everything (778305, 2017-21, PI & Coordinator)
    [H2020] Mo-IoNT - Molecular Internet-of-Nano-Things (792799, 2018-20, PI)
    [H2020] GreenML5G - Green Machine Learning for 5G & Beyond Resource 
    Optimisation (891221, 2020-22, PI)


  • Physical - Sensing and Digital Twin of Critical Systems - We are experiencing a data-centric engineering revolution where vast volumes of sensor data can inform digital twin models. However, how much data is really needed and what is the connection between complex engineering physics and data science is not established. This is particularly an issue in cyberphysical coupled and highly connected large-scale national critical infrastructure systems. Our research focus on:
    1) Resilience prediction of connected systems, coupling engineering ODE/PDE physics with graph theory (see our EPSRC funded work on resilience of networked dynamics in Nat. Sci. Rep., and on modeling connected geomagnetic storms in Nature Communications)
    2) Data science of networked systems, developing a connection between engineering physics and Graph Fourier transforms as a minimum data sampling theory (see our IEEE Trans Signal Processing paper on GFTs to sample connected nonlinear dynamicsand its neural network version to water distribution networks)

    Key Grants: [EPSRC] CoTRE - Complexity Twin for Resilient Ecosystems (EP/R041725/1, 2018-20, PI)
    [Turing LRF] CHANCE - Coupled Human and Natural Critical Ecosystems (2017-21, PI)


  • Social - Human Response to a Changing Environment - Our society is experiencing change faster than ever both from technology and from the natural environment. This has induced political violence and caused mass migration. Our work focus on developing social physical models for conflict and migration at the global scale:
    1) Global conflict prediction using networked socio-physical models trained on historical data was funded by UK government via the Alan Turing Institute under project GUARD​ with Sir Alan Wilson. This has led to BBC interviews and now is a commercial software within the DSTL/MOD community. Our thought process on AI and conflict prediction can be found here in Nature, and our recent socio-physics informed machine learning models on excitation processes can be found in IJCAI.
    2) Smart cities and living can provide the critical benefits for our cities to flourish or lead to more problems. My research focus on the data science of technology usage in cities from the first case study of 4G energy consumption in London to financial access disparities (evidence given to UK Treasury), to examining how historical technology revolutions changed our urbanisation trends.

    Key Grants: [EPSRC] Centre for Doctoral Training in Urban Science and Progress (EP/L016400/1, 2014-22, CI)
    [EPSRC] Patterns of City Formation & Development (EP/N509796/1, 2018-22, PI)
    [USAF] Networked Social Influence and Acceptance in a New Age of Crises (FA8655-20-1-7031, 2020-23, PI)
    [Turing D&S] GUARD - Global Urban Analytics for Resilient Defence (2017-19, extension: 2019-21, PI)

Background - I obtained MEng, MA, and PhD degrees from the University of Cambridge. From 2012 to 2019, I built an award-winning team at the University of Warwick. Since 2019, I joined Cranfield University (Engineering: REF UK 2021 #7, Mech-Aero Eng: QS world 2022 #27) at the Centre for Autonomous and Cyber-Physical Systems [Link]. Here, I lead the Human Machine Intelligence Research Group, where I work closely with the new £65m Digital Aviation Research and Technology Centre (DARTeC). 

Track-record - I have been PI on over £5.8m and an investigator on over £18m of research funding. ​I was a Turing Fellow at the Alan Turing Institute from 2017 and remain a researcher there. I have published over 120 journal papers (total IF 550+) and 70+ IEEE/ACM conference papers, with 5000+ citations (h-index 34). Key papers published include: a Nature commentary, Nature Comm., Nature Comp. Science, a top 10% cited PLOS One paper, and several cover issues in Royal Society and IEEE Transactions. I currently serve as editor on 3 IEEE & 1 Royal Society journals and reviewer for EPSRC (full college), ESRC, MRC, UKRI FF, NSF, NSERC, H2020, Royal Society, Leverhulme, and other international grant awarding bodies. I have been an IET Innovation Award winner (2015) and been a runner-up in the Bell Labs Prize three times.

Personal Life - I have worked in an UNHCR refugee camp in N. Africa, been part of the victorious Cambridge Varsity archery team, solo climbed two of the highest sub-continent peaks (Toubkal-Atlas Mt, Jade Mt), served as badminton captains at both Cambridge (Fitz) and Sheffield Universities, and completed both the London (2007) and full Sahara (2010) full marathons. My personal interests lie in fitness, world peace and refugee safety.

Key Papers
Screenshot 2021-05-18 at 07.27.06.png
Screenshot 2021-05-18 at 07.29.53.png
Screenshot 2021-05-18 at 07.58.55.png
Screenshot 2021-05-18 at 07.27.45.png
Screenshot 2021-05-17 at 20.19.50.png
Screenshot 2021-05-17 at 19.58.05.png
Screenshot 2021-05-27 at 09.13.22.png
Screenshot 2021-05-27 at 09.01.40.png
Screenshot 2021-05-27 at 08.58.07.png

"Random Sketch Learning for Deep Neural Networks in Edge Computing,"
B. Li, P. Chen, H. Liu, W. Guo, X. Cao, J. Du, C. Zhao, J. Zhang,
Nature Computational Science, vol.1(3), 2021 [Link] [Commentary]

"Edge implementation of DNNs require extraordinary computational power. Current compression post-training techniques are not sufficient. Here we present the first compress-whilst-training tiny AI saving up to 10x energy and 180x computation power."  

"Retool AI to Forecast and Limit Wars,"
W. Guo, K. Gleditsch, A. Wilson,
Nature, vol. 562, 331-333, Oct 2018 [Link]

"Conflict causes long term damage to the socioeconomic fabric of civilization and hinders the UN Development Goals. Here we show AI without domain knowledge has limited applicability and cross-disciplinary research is needed to improve forecasting in rural and sparse data areas."   [Funded by Alan Turing Institute D&S Program]

"Scalable Partial Explainability in Neural Networks via Flexible Activation Functions,"
S. Sun, C. Li, Z. Wei, A. Tsourdos, W. Guo,
AAAI Conference on Artificial Intelligence, Feb 2021 [Link]

"Explaining the role of activation functions via flexible GPs with control points is useful for interpreting features without an initial bias function. Here we demonstrate a scalable understanding and compression of neural networks for detecting fake currency." 

"Revealing the Excitation Causality between Climate and Political Violence via a Neural Forward-Intensity Poisson Process,"
S. Sun, B. Jin, Z. Wei, W. Guo,
International Joint Congress on AI (IJCAI), July 2022 [Link]
"Conflict drivers often excite a social transformation process which leads to violence, but further climate effects do not contribute to further violence. We use a neural forward-intensity Poisson process to model the nonlinear excitation causal mechanism. Our 20 year results reveal an excitation-based causal link between climate events and global conflict, cross-validated against qualitative climate vulnerability indices."  [Funded by Turing D&S]

"Neural Network Approximation of Graph Fourier Transforms for Sparse Sampling of Networked Flow Dynamics,"
A. Pagani, Z. Wei, R. Silva, W. Guo,
ACM Transactions on Internet Technology, to appear, Jun 2021

"Compressing networked non-linear dynamical systems is crucial to sensing complex infrastructures. Here we train deep learning to both compress (using Graph Fourier Transform as a training basis) networked non-linear dynamics of a water distribution network, and then to perform sub-GFT inference with partial data. We show neural networks can achieve acceptable noisy inference with only a small set of sensors."  [Funded by EPSRC CoTRE and Alan Turing Institute DCE Program] 

"Network Community Structure of Substorms using SuperMAG Magnetometers,"
L. Orr, S. Chapman, J. Gjerloev, W. Guo,
Nature Communications, vol.12, 1842, 2021 [Link
"Geomagnetic storms can disrupt electronic and GNSS systems, but their formation is not well understood. Here we use network science to show that small sub-storms are in fact interconnected and form the early phase of larger geomagnetic storms."   [Funded by Alan Turing Institute & USAFOSR & EPSRC CDT in Complexity]

"Node-Level Resilience Loss in Dynamic Complex Networks,"
G. Moutsinas, W. Guo,
Nature Scientific Reports, vol.10, 2020 [Link]
"Complex engineering and ecological systems derive sophisticated capabilities by networking together simple components. Yet, the multi-scale resilience of networked non-linear systems is not well understood. Here, we show a sequential heterogeneous mean field approach that can rapidly converge on multi-scale resilience estimation as a function of both the component level non-linear dynamics and the graph properties of the network. In particular, we show how higher order effects can cascade along the network, ending in rupture at the end of a network, yet trace causes to the beginning of a health network."  [Funded by EPSRC CoTRE]

"Sampling and Inference of Networked Dynamics using Log-Koopman Nonlinear Graph Fourier Transform,"
Z. Wei, B. Li, SC. Sun, W. Guo,
IEEE Transactions on Signal Processing, Nov 2020 [Link

"Compressing networked non-linear dynamical systems is crucial to sensing complex infrastructures. Here we combine GFTs with a novel log-Koopman operator to efficiently lineaepize large-scale connected dynamics and compress to a spatially invariant sensing solution."  [Funded by EPSRC CoTRE and Alan Turing Institute DCE Program] 


"Molecular Physical Layer for 6G in Wave-Denied Environments," 
W. Guo, M. Abbaszadeh, L. Lin, J .Charmet, P. Thomas, Z. Wei, B. Li, C. Zhao,
IEEE Communications Magazine, vol.59(5), May 2021 [Link]

"Review of 8 years of funded research in molecular communications from theory to experimentation for future 6G bio-air-interface. Our work examines the information theory of fluid dynamic channels for a wide range of micro and macro scales. We use tracer molecules ignited by PLIF and other laser techniques to image the propagation of information molecules in different fluid dynamic channels with diverse obstacles and understand the impact on information capacity."  [Funded by H2020 Mo-IoNT, DSTL MEDE, USAF MolCom, Royal Society-NSFC]


"Explainable Artificial Intelligence (XAI) for 6G: Improving Trust between Human and Machine,"
W. Guo,
IEEE Communications Magazine, vol.58(6), 2020 [Link]
"Review of the importance of Explainable AI for future communication systems that are vital to connected autonomy and safety critical areas. Our work examines both physical layer and MAC layer protocols and intelligence. Examples include explaining neural network power allocation in multiple access systems using meta-symbolic mapping."  [Funded by H2020 DAWN4IoE]

"Probabilistic Stability of Traffic Load Balancing on Wireless Complex Networks,"
G. Moutsinas, W. Guo,
IEEE Systems Journal, 2019 [Link]

"Seminal work to prove that load balancing stability in any network is independent of the dynamics and topology, but dependent on the accuracy of the load information. Here we use a wireless network as an example, but can be extended to other systems, and provided the self-dynamics of stations and balancing dynamics between stations are reasonable load balancing ODEs of any form, we show it is always stable for any graph topology using stochastic geometry. We further show through Gershgorin circles that the eigenvalues are bounded in stable regime only if the data accuracy that informs the load balancing has a variance that is smaller than the variance of noise."  [Funded by EPSRC CoTRE]