Staff Profiles

I have an extensive teaching experience at university level in three different countries, first as a Teaching Assistant (TA) in Padua and Milan, later as a TA at the University of California at Los Angeles, and eventually as the Module Leader for a number of Undergraduate and Postgraduate courses at Oxford Brookes University, UK. At Brookes I have supervised several Master's students' dissertations and third year’s final projects, and I regularly mentor a number of undergraduate students as their Academic Adviser. As the Subject Coordinator of the MSc in Computer Vision I was responsible for embedding the Course within the research activities of the AI and Vision group, and I personally designed the four core modules of the course.

Taught courses at PG level: P00990 “Research and Scholarship Methods" (2013-14), P00405 “Mathematical Methods for Computer Vision" (2013-17), P00406 “Machine Learning" (2013-17), P00408 “Advanced Computer Vision" (2015-16), P08821 "Advanced Machine Learning", all as the Module leader, plus P00702 “Cyber security and the web” as co-teacher. At UG level: U08781 “Machine Vision" (2008-2019), U08280 “Advanced Artificial Intelligence" (2009-2013), U08884 “Image Technology" (2009-2015) as the Module Leader, plus U08027 “Current research” as co-teacher.

Modules taught

Undergraduate modules:

U08781 “Machine Vision" (module leader)

Postgraduate modules

P08821 “Advanced Machine Learning” (module leader)

Supervision

Fellows

Reza Alitappeh (March 2019-now), ECM Fellow in AI for autonomous driving.

Postdoctoral researchers  

Suman Saha (Dec 2017-Jul 2018), now postdoc at ETH Zurich; Ruomei Yan (Sep 2017-Dec 2017), now with ARM Semiconductors; Ahmed Samet (Mar 2016 - onward), Project title: “New robust foundations of statistical machine learning”; Wenjuan Gong (Feb 2013 - Jan 2014), funded by EPSRC First Grant, Project title: “Tensorial modeling of dynamical systems for gait and activity recognition" (now Lecturer at China University of Petroleum, Qingdao, Shandong, China).

KTP Associates

Neha Bhargava (Apr 2019-21), funded by the KTP with Createc Technologies.

Ruomei Yan (Sept 2015 - Nov 2017), funded by Meta Vision KTP, Project title: “Sensing a good weld: new applications in computer vision" (formerly a Lecturer at Shanghai Chan-Mai University, the 3^rd best Chinese university).

Ph.D. students (as Director of Studies, unless otherwise indicated) 

Michael Sapienza (Oct 2011 - Oct 2014), Thesis title: “Recognising and localising human actions" (now with Samsung Research, formerly postdoc at Oxford University, Department of Engineering Science); Vinhav Vineet (January 2014 - October 2014), second supervisor, thesis title: “Recognition, Reorganisation, Reconstruction and Reinteraction for Scene Understanding" (now postdoc at Stanford University); Min Han Lee (September 2014 - August 2015), Topic: “Action recognition from unconstrained videos"; Suman Saha (September 2014 - present), Topic: “Uncertainty in Computer Vision", Winner of best MSc dissertation at University of Bedfordshire and best reading group competition at the 2015 International Summer School of Computer Vision (ICVSS 2015); Gurkirt Singh (September 2015 - present), Topic: “Real-time Action Recognition for Human-Robot Interaction", currently intern at Borealis AI, Vancouver. Wojtek Buczynski (Oct 2018 - now), based at Cambridge University, second supervisor (DoS Barbara Sahakian).

Visiting students

Rocco de Rosa (January 2014 - July 2014), EU Erasmus Training Programme (now Data Science Manager at Rank Group, formerly a postdoc at University of Rome “La Sapienza"); Serdar Buyukkanli (June 2014 - September 2014), funded by EU Erasmus Training Programme; Brenda Romino (February 2016 - August 2016), funded by University of Naples; Manuele Di Maio (March 2017 - September 2017), funded by Erasmus+; Andrea Morelli (January 2017 - May 2017), funded by Erasmus+; Shashwat Shukla, Oxford Brookes - IIT Bombay exchange programme. Santanu Rathod (May-Jul 2018), Brookes - IIT programme; Valentina Fontana (Mar-Sep 2018), Erasmus+ trainee from Naples’ Federico II; Giacomo De Rossi (Sep-Nov 2018), Univ of Verona.

MSc students

Brad Lishman (2012), Stephane Bourgeois (2014), Jonathan Pound (2014), Paula Rocafull (2015), Ben Guy (2016), Kurt DeGiorgio (2016), Misbah Munir (2016-17), Stavros Gasparis (2017), K. Parshotam (2017), Stephen Akkrigg (2018), Francis Kaping’A (2018).

External examiner roles

On March 30 2012 I was external examiner for the Ph.D. candidate Andrea Argentini, University of Trento, Italy (title of the thesis: “Ranking Aggregation Based on Belief Function Theory”). I currently am external examiner for Zhenghua Xu, Department of Computer Science, Oxford University, and Timber Kerkvliet, VU University Amsterdam. I have been internal examiner for Ph.D. candidates Chris Russell (title: “Higher-order inference for vision problems”) and Paul Sturgess (April 2016).

As the founder and head of Brookes’ Visual Artificial Intelligence research group, I have been conducting work at the current boundaries of human action recognition. My group has built in just a few years a leadership position in the field of deep learning for real-time action detection, localisation and recognition, which has led to be best detection accuracies to date and the only system able to localise multiple actions on the image plane in (better than) real time. We are now shifting towards work at the current boundaries of visual AI, such as: (i) the design of new deep learning architecture able to regress whole action tubes in real time; (ii) structured-output DNs with as output part-based discriminative models, (iii) deep neural video captioning incorporating attention models and prior logical knowledge, & (iv) the creation of a theory of mind for visual AIs. In the past I have proposed spectral embedding techniques for unsupervised 3D segmentation and matching which have been rather highly cited in the field (see Google metrics). The group has heavily invested in action and activity recognition via discriminative part-based models, in partnership with Oxford’s Torr Vision group, generating an IJCV paper and prizes at MLVR and BMVC, and published results on metric learning for generative dynamical models (cfr. PAMI 2014 article).

I am a recognised leader in the field of uncertainty theory and belief functions. My reputation comes from the formulation of a geometric approach to uncertainty in which probabilities, possibilities, belief measures and random sets are represented and analysed by geometric means. This work is going to be published as a Springer monograph “The geometry of uncertainty” in 2017, and was published as a separate monograph by Lambert in 2014. My work concerns all the mathematical properties of non-additive probabilities and their application to decision making under partial or missing data, including: the generalisation of the law of total probability, the notions of upper and lower likelihoods, and that of generalised random variables.

Within machine learning my work is directed at understanding the mathematics of deep learning, providing new robust foundations for statistical learning theory, and developing novel tools based on the theory of random sets, in particular generalisation of the logistic regression frameworks and of max-entropy classifiers. I also worked on manifold learning for generative dynamical models, and the generalization of bilinear classifiers to the tensorial case in order to model multiple nuisance factors (see EPSRC First Grant). I am part of research consortia which aim at applying machine learning to human-robot interaction (the creation of emotional avatars), factory robotics (the coordination of fleets of automated forklifts), surgical robotics (robotic assistant surgeon arms) and healthcare (home monitoring and the early diagnosis of dementia from accelerometer and video data captured via wearable devices, cfr. a recent EPSRC bid and a joint Gait & Posture paper with Prof Dawes).

Research group membership

Visual Artificial Intelligence Laboratory (Director): http://cms.brookes.ac.uk/staff/FabioCuzzolin 

Torr Vision Group, Oxford University (Associate member): www.robots.ox.ac.uk/~tvg/

Research grants and awards

To date I attracted external fundingfor a total of circa £1,300,000 (not fully incorporating the €4.3M Horizon 2020 project SARAS, which I co-wrote with the Coordinator Riccardo Muradore), and internally-funded studentships for an equivalent amount of £216,000. As head of a research unit I annually receive £28,000 in QR money from the Department, as a result of the group's REF 2014 performance. In 2014-2015 four other bids reached the final stage or scored very highly but were eventually not funded, for a total of £1,615,000.

Several grant applications are (as of November 2017) pending for around £1-2M, including a Leverhulme Trust Research Grants at the final, full proposal stage, a Horizon 2020 COST action, a Horizon 2020 FET bid, the Expanding Excellence in England AGILE bid. A host of other applications are in the making, including two Engineering and Physical Sciences (EPSRC) ones, a H2020 Factory of the Future (FoF) project, and two with Innovate UK led by Cortexica Vision Systems and AnyVision. More details below.

Past and current funding

• Oxford Brookes University - Central Research Funding (CRF). Uncertainty in Computer Vision (Dec 2008). Travel money for a total of £3,000.
• Oxford Brookes University - “Intelligent Transport Systems" doctoral training programme. Multi Sensor Fusion for Simultaneous Localization and Mapping on Autonomous Vehicles  (Jan 2011). Own role: Director of Studies. Period: September 2011 - October 2014. One PhD Studentship.
• EPSRC - First Grant Scheme. Tensorial modeling of dynamical systems for gait and activity recognition (Feb 2011). Own role: Principal Investigator. Budget: £122,000. Period: July 2011 - January 2014. Rated 6/6, 6/6, 5/6, 6/6. The project has generated three articles on PAMI, IJCV and IEEE TFS, and a number of follow-up grant bids.
• Oxford Brookes University - Faculty of Technology: Next 10 Programme Award (Oct 2012). Own role: Director of Studies. One PhD Studentship. Period: September 2014 - October 2017.
• Onera - Elsevier – Int. Society of Information Fusion (ISIF): Sponsorship of BELIEF 2014 (2014). £7,000.
• Oxford Brookes 150th Anniversary Scholarship: Online action recognition for human-robot interaction (July 2014). Own role: Director of studies. Personnel: one PhD studentship. Period: September 2015-October 2018.
• Innovate UK: Meta Vision Knowledge Transfer Partnership (Apr 2015). Own role: Academic supervisor. Period: September 2015 - August 2017. Budget: £160,000. Personnel: one KTP associate. In Partnership with Meta Vision LTD.
• NVIDIA: Hardware Grant Request (Oct 2015). Donation: one Titan X GPU card to support the group's work on online action recognition (£650).
• Oxford Brookes - IIT Bombay internship scheme: Action detection and recognition from videos (Jan 2017). Budget: £3,500. Role: Supervisor.

• Horizon 2020, Call ICT-27-2017: SARAS - Smart Autonomous Robotic Assistant Surgeon (Aug 10 2017). Duration: 3 years. Coordinator: U. Verona, Italy. Own role: Scientific Officer (SO) and WP Leader. Together with Coordinator Riccardo Muradore I was the main contributor to the success of this bid. In particular, I have single-handedly rewritten both part 2 (impact) and part 3 (management), without which the project would never have been funded. Budget: €4,315,640 (own share: €596,073).

• Innovate UK - Knowledge Transfer Partnership with Createc (August 2018). Own role: Academic supervisor and Lead Academic. Budget: £190,000. Duration: 2 years. Personnel: one KTP associate.

• Oxford Brookes University, School of Engineering, Computing and Mathematics - Fellowship in AI for autonomous driving (August 2018). Own role: PI. Budget: £100,000. Duration: 2 years.
• Huawei Technologies - Video analysis and activity recognition (April 2019). Own role: PI. Budget: £280,000. Duration: 3 years. Personnel: one postdoctoral researcher, one PhD student.
• UKIERI - Some novel paradigms for analyzing human actions in complex videos (March 2019). Own role: PI for the UK side. Budget: £42,000. Co-PI Prof Subhasis Chaudhuri, Director of IIT Bombay.

Awards

• 2017 CVPR Charades challenge, 2^nd place (with student G. Singh), June 2017 (1st place won by Google Deepmind’s TeamKinetics).

• 2016 CVPR ActivityNet action detection challenge, 2^nd place (with student G. Singh). Jun 2016.

• Next 10 Award - Oxford Brookes University - Faculty of Technology (Oct 2012). Research accelerator programme, awarded to the top emerging researchers in the Faculty. 

• Outstanding Reviewer Award - British Machine Vision Conference - BMVC 2012 (Sep 2012).
• Short-listed for the Best Paper Award - British Machine Vision Conference (BMVC 2012). For the paper: “Learning discriminative space-time actions from weakly labelled videos".
• Best Poster Prize - INRIA Visual Recognition and Machine Learning Summer School (VRML 2012), July 2012. For the poster: “Learning discriminative space-time actions from weakly labelled videos".
• Best Poster Award - International Symposium on Imprecise Probabilities Theories and Applications (ISIPTA’11), July 2011. For the poster: “Geometric conditional belief functions in the belief space".
• Short-listed for the Best Paper Award - ECSQARU 2011, Jun 2011. For the paper “On consistent approximations of belief functions in the mass space".
• Best Paper Award - Pacific Rim International Conference on Artificial Intelligence (PRICAI’08), Dec 2008. For the paper: “Alternative formulations of the theory of evidence based on basic plausibility & commonality assignments. 
• Marie Curie fellowship (Sep 2006), in partnership with INRIA Rhone-Alpes, Perception group.

In addition, my PhD student Suman Saha has won the reading group prize at ICVSS 2015, the International Computer Vision Summer School. My MSc student Misbah Munir has won in February 2017 the OBSEA (the Oxford Brookes Social Entrepreneur Awards) Try It Award to fund a proof of concept of her work on deep learning for video captioning.

Research projects

Computer Vision

The AI and Vision group, in close collaboration with Oxford University's Torr Vision Group, has a multi-year experience in hot computer vision topics such as action, gesture and activity recognition, pose estimation, segmentation and matching of articulated bodies, voxelset analysis, video retrieval. Machine learning techniques employed range from deep learning and convolutional neural networks, to hidden Markov models, metric learning, dimensionality reduction, and discriminative part-based models.

Machine Learning

The group is active in machine learning (in particular metric learning for dynamical models, weakly supervised classification, imprecise dynamical models), and its application to problems such as big data, gait and daily activity analysis for dementia diagnosis and monitoring, vehicle classification via inductive loops, and activity localisation and recognition.

Uncertainty Theory

Prof Cuzzolin is a world expert in the theory of belief functions, having published two monographs on the topic. His main contribution is a geometric approach to uncertainty theory in which every measure can be represented as a point of an appropriate convex space. The group is active on a number of topics in this field, including: probability and possibility transformation for efficient computation, decision making, the algebra of decision spaces (frames of discernment), and the total belief theorem.

Artificial Intelligence

Artificial intelligence is already part of our lives. Smart cars will engage our roads in less than ten years' time; shops with no checkout, which automatically recognise customers and what they purchase, are already open for business. But to enable machines to deal with uncertainty, we must fundamentally change the way machines learn from the data they observe so that they will be able to cope with situations they have never encountered in the safest possible way. Interacting naturally with human beings and their complex environments will only be possible if machines are able to put themselves in people's shoes: in other words, to read our minds.

Robotics

The AI and Vision group collaborates very closely with the Cognitive Robotics group, led by Professor Nigel Crook and Dr Matthias Rolf. In particular, we are interested in human-robot interaction, EEG classification coupled with humanoid robots for the creation of emotional robot avatars, and robot assistants for laparoscopic surgery.
The two groups are involved in the Intelligent Transport Systems doctoral training programme, and collaborate with the Autonomous Driving research group led by Dr Andrew Bradley.

E-health

We also closely collaborate with Prof Helen Dawes and her MOReS centre on the application of AI, machine learning and vision to healthcare, including: the early diagnosis of dementia, the diagnosis of diabetes from gait signatures, and the monitoring of people with prolonged disorder of consciousness using multimodal deep learning.

For more information please consult the Visual AI Laboratory web site (see below).

Further information

http://cms.brookes.ac.uk/staff/FabioCuzzolin/

Books

• Cuzzolin F, The Geometry of Uncertainty - The Geometry of Imprecise Probabilities, Springer International Publishing (2021)
  ISBN: 9783030631529 eISBN: 9783030631536
  Abstract

  The principal aim of this book is to introduce to the widest possible audience an original view of belief calculus and uncertainty theory. In this geometric approach to uncertainty, uncertainty measures can be seen as points of a suitably complex geometric space, and manipulated in that space, for example, combined or conditioned. 

  In the chapters in Part I, Theories of Uncertainty, the author offers an extensive recapitulation of the state of the art in the mathematics of uncertainty. This part of the book contains the most comprehensive summary to date of the whole of belief theory, with Chap. 4 outlining for the first time, and in a logical order, all the steps of the reasoning chain associated with modelling uncertainty using belief functions, in an attempt to provide a self-contained manual for the working scientist. In addition, the book proposes in Chap. 5 what is possibly the most detailed compendium available of all theories of uncertainty. Part II, The Geometry of Uncertainty, is the core of this book, as it introduces the author’s own geometric approach to uncertainty theory, starting with the geometry of belief functions: Chap. 7 studies the geometry of the space of belief functions, or belief space, both in terms of a simplex and in terms of its recursive bundle structure; Chap. 8 extends the analysis to Dempster’s rule of combination, introducing the notion of a conditional subspace and outlining a simple geometric construction for Dempster’s sum; Chap. 9 delves into the combinatorial properties of plausibility and commonality functions, as equivalent representations of the evidence carried by a belief function; then Chap. 10 starts extending the applicability of the geometric approach to other uncertainty measures, focusing in particular on possibility measures (consonant belief functions) and the related notion of a consistent belief function. The chapters in Part III, Geometric Interplays, are concerned with the interplay of uncertainty measures of different kinds, and the geometry of their relationship, with a particular focus on the approximation problem. Part IV, Geometric Reasoning, examines the application of the geometric approach to the various elements of the reasoning chain illustrated in Chap. 4, in particular conditioning and decision making. Part V concludes the book by outlining a future, complete statistical theory of random sets, future extensions of the geometric approach, and identifying high-impact applications to climate change, machine learning and artificial intelligence. 

  The book is suitable for researchers in artificial intelligence, statistics, and applied science engaged with theories of uncertainty. The book is supported with the most comprehensive bibliography on belief and uncertainty theory. -- Provided by publisher.

  Website 
• Destercke S, Denoeux T, Cuzzolin F, Martin A, Ed., Belief Functions: Theory and Applications (BELIEF, 5th International Conference. Proceedings), Springer International Publishing (2018)
  eISSN: 1611-3349 ISBN: 9783319993829 eISBN: 9783319993836
  Abstract

  This book constitutes the refereed proceedings of the 5th International Conference on Belief Functions, BELIEF 2018, held in Compiègne, France, in September 2018. The 33 revised regular papers presented in this book were carefully selected and reviewed from 73 submissions. The papers were solicited on theoretical aspects (including for example statistical inference, mathematical foundations, continuous belief functions) as well as on applications in various areas including classification, statistics, data fusion, network analysis and intelligent vehicles.

  Website 
• Cuzzolin F, Ed., Belief Functions: Theory and Applications, Springer International Publishing (2014)
  ISBN: 978-3-319-11191-9
  Website 
• Cuzzolin F, Visions of a generalized probability theory, Lambert Academic Publishing (2014)
  ISBN: 978-3-659-13175-2
  

Journal articles

• Cuzzolin F, Morelli A, Cîrstea B, Sahakian BJ, 'Knowing me, knowing you: theory of mind in AI'
  Psychological Medicine 50 (7) (2020) pp.1057-1061
  ISSN: 0033-2917 eISSN: 1469-8978
  Abstract

  Artificial intelligence has dramatically changed the world as we know it, but is yet to fully embrace ‘hot’ cognition, i.e., the way an intelligent being's thinking is affected by their emotional state. Artificial intelligence encompassing hot cognition will not only usher in enhanced machine-human interactions, but will also promote a much needed ethical approach. Theory of Mind, the ability of the human mind to attribute mental states to others, is a key component of hot cognition. To endow machines with (limited) Theory of Mind capabilities, computer scientists will need to work closely with psychiatrists, psychologists and neuroscientists. They will need to develop new models, but also to formally define what problems need to be solved and how the results should be assessed.
  

  Website 
• Leporini A et al, 'Technical and Functional Validation of a Teleoperated Multirobots Platform for Minimally Invasive Surgery'
  IEEE Transactions on Medical Robotics and Bionics 2 (2) (2020) pp.148-156
  ISSN: 2576-3202 eISSN: 2576-3202
  Abstract

  Nowadays Robotic assisted Minimally Invasive Surgeries (R-MIS) are the elective procedures for treating highly accurate and scarcely invasive pathologies, thanks to their ability to empower surgeons’ dexterity and skills. The research on new Multi-Robots Surgery (MRS) platform is cardinal to the development of a new SARAS surgical robotic platform, which aims at carrying out autonomously the assistants tasks during RMIS procedures. In this work, we will present the SARAS MRS platform validation protocol, framed in order to assess: (i) its technical performances in purely dexterity exercises, and (ii) its functional performances. The results obtained show a prototype able to put the users in the condition of accomplishing the tasks requested (both dexterity- and surgical-related), even with reasonably lower performances respect to the industrial standard. The main aspects on which further improvements are needed result to be the stability of the end effectors, the depth perception and the vision systems, to be enriched with dedicated virtual fixtures. The SARAS’ aim is to reduce the main surgeon’s workload through the automation of assistive tasks which would benefit both surgeons and patients by facilitating the surgery and reducing the operation time.
  

  Website 
• Liu Z, Liu Y, Dezert J, Cuzzolin F, 'Evidence combination based on credal belief redistribution for pattern classification'
  IEEE Transactions on Fuzzy Systems 28 (4) (2019) pp.618-631
  ISSN: 1063-6706
  Abstract

  Evidence theory, also called belief functions theory, provides an efficient tool to represent and combine uncertain information for pattern classification. Evidence combination can be interpreted, in some applications, as classifier fusion. The sources of evidence corresponding to multiple classifiers usually exhibit different classification qualities, and they are often discounted using different weights before combination. In order to achieve the best possible fusion performance, a new Credal Belief Redistribution (CBR) method is proposed to revise such evidence. The rationale of CBR consists in transferring belief from one class not just to other classes but also to the associated disjunctions of classes (i.e., meta-classes). As classification accuracy for different objects in a given classifier can also vary, the evidence is revised according to prior knowledge mined from its training neighbors. If the selected neighbors are relatively close to the evidence, a large amount of belief will be discounted for redistribution. Otherwise, only a small fraction of belief will enter the redistribution procedure. An imprecision matrix estimated based on these neighbors is employed to specifically redistribute the discounted beliefs. This matrix expresses the likelihood of misclassification (i.e., the probability of a test pattern belonging to a class different from the one assigned to it by the classifier). In CBR, the discounted beliefs are divided into two parts. One part is transferred between singleton classes, whereas the other is cautiously committed to the associated meta-classes. By doing this, one can efficiently reduce the chance of misclassification by modeling partial imprecision. The multiple revised pieces of evidence are finally combined by Dempster-Shafer rule to reduce uncertainty and further improve classification accuracy. The effectiveness of CBR is extensively validated on several real datasets from the UCI repository, and critically compared with that of other related fusion methods.

  Website 
• Gong W, Cuzzolin F, 'A belief-theoretical approach to example-based pose estimation'
  IEEE Transactions on Fuzzy Systems PP (99) (2017) pp.1-14
  ISSN: 1063-6706
  Abstract In example-based human pose estimation, the configuration of an evolving object is sought given visual evidence, having to rely uniquely on a set of sample images. We assume here that, at each time instant of a training session, a number of feature measurements is extracted from the available images, while ground truth is provided in the form of the true object pose. In this scenario, a sensible approach consists in learning maps from features to poses, using the information provided by the training set. In particular, multi-valued mappings linking feature values to set of training poses can be constructed. To this purpose we propose a Belief Modeling Regression (BMR) approach in which a probability measure on any individual feature space maps to a convex set of probabilities on the set of training poses, in a form of a belief function. Given a test image, its feature measurements translate into a collection of belief functions on the set of training poses which, when combined, yield there an entire family of probability distributions. From the latter either a single central pose estimate or a set of extremal ones can be computed, together with a measure of how reliable the estimate is. Contrarily to other competing models, in BMR the sparsity of the training samples can be taken into account to model the level of uncertainty associated with these estimates. We illustrate BMR’s performance in an application to human pose recovery, showing how it outperforms our implementation of both Relevant Vector Machine and Gaussian Process Regression. Finally, we discuss motivation and advantages of the proposed approach with respect to its most direct competitors.Website 
• De Rosa R, Gori I, Cuzzolin F, Cesa-Bianchi N, 'Active Incremental Recognition of Human Activities in a Streaming Context'
  Pattern Recognition Letters 99 (2017) pp.48-56
  ISSN: 0167-8655
  Abstract Recognising human activities from streaming sources poses unique challenges to learning algorithms. Predictive models need to be scalable, incrementally trainable, and must remain bounded in size even when the data stream is arbitrarily long. In order to achieve high accuracy even in complex and dynamic environments methods should be also nonparametric, i.e., their structure should adapt in response to the incoming data. Furthermore, as tuning is problematic in a streaming setting, suitable approaches should be parameterless (as initially tuned parameter values may not prove optimal for future streams). Here, we present an approach to the recognition of human actions from streaming data which meets all these requirements by: (1) incrementally learning a model which adaptively covers the feature space with simple and local classifiers; (2) employing an active learning strategy to reduce annotation requests; (3) achieving good accuracy within a fixed model size. Although in this work we focus on human activity recognition, our approach is completely independent from the feature extraction and can deal with any supervised matrix (set of feature vectors). Hence, it can be adapted to a wide range of applications (e.g., speech recognition, image classification, object recognition, pose recognition, and image matching). Extensive experiments on standard benchmarks show that our approach is competitive with state-of-the-art non-incremental methods, while outperforming the existing active incremental baselines.Website 
• Cuzzolin F, Sapienza M, Esser P, Saha S, Franssen M, Collett J, Dawes H, 'Metric learning for Parkinsonian identification from IMU gait measurements'
  Gait and Posture 54 (May 2017) (2017) pp.127-132
  ISSN: 0966-6362 eISSN: 1879-2219
  Abstract Diagnosis of people with mild Parkinson’s symptoms is difficult. Nevertheless, variations in gait pattern can be utilised to this purpose, when measured via Inertial Measurement Units (IMUs). Human gait, however, possesses a high degree of variability across individuals, and is subject to numerous nuisance factors. Therefore, off-the-shelf Machine Learning techniques may fail to classify it with the accuracy required in clinical trials. In this paper we propose a novel framework in which IMU gait measurement sequences sampled during a 10 metre walk are first encoded as hidden Markov models (HMMs) to extract their dynamics and provide a fixed-length representation. Given sufficient training samples, the distance between HMMs which optimises classification performance is learned and employed in a classical Nearest Neighbour classifier. Our tests demonstrate how this technique achieves accuracy of 85.51% over a 156 people with Parkinson’s with a representative range of severity and 424 typically developed adults, which is the top performance achieved so far over a cohort of such size, based on single measurement outcomes. The method displays the potential for further improvement and a wider application to distinguish other conditions.
   
  Website 
• Cuzzolin F, 'Belief functions: Theory and applications (BELIEF 2014)'
  International Journal of Approximate Reasoning 72 (2016) pp.1-3
  ISSN: 0888-613X
  Website 
• Cuzzolin F, 'Lp consonant approximations of belief functions'
  IEEE Transactions on Fuzzy Systems 22 (2) (2014) pp.420-436
  ISSN: 1063-6706
  Abstract

  In this paper, we solve the problem of approximating a belief measure with a necessity measure or “consonant belief function” in a geometric framework. Consonant belief functions form a simplicial complex in both the space of all belief functions and the space of all mass vectors: Partial approximations are first sought in each component of the complex, while global solutions are selected among them. As a first step in this line of study, we seek here approximations that minimize the mass space can be interpreted in terms of mass redistribution, while approximations in the belief space generalize the maximal outer consonant approximation.We compare them with each other and with other classical approximations and illustrate them with the help of a running example.

  Website 
• Antonucci A, De Rosa R, Giusti A, Cuzzolin F, 'Robust classification of multivariate time series by imprecise hidden Markov models'
  International Journal of Approximate Reasoning 56 (Part B) (2014) pp.249-263
  ISSN: 0888-613X
  Abstract

  A novel technique to classify time series with imprecise hidden Markov models is presented. The learning of these models is achieved by coupling the EM algorithm with the imprecise Dirichlet model. In the stationarity limit, each model corresponds to an imprecise mixture of Gaussian densities, this reducing the problem to the classification of static, imprecise-probabilistic, information. Two classifiers, one based on the expected value of the mixture, the other on the Bhattacharyya distance between pairs of mixtures, are developed. The computation of the bounds of these descriptors with respect to the imprecise quantification of the parameters is reduced to, respectively, linear and quadratic optimization tasks, and hence efficiently solved. Classification is performed by extending the k-nearest neighbors approach to interval-valued data. The classifiers are credal, meaning that multiple class labels can be returned in the output. Experiments on benchmark datasets for computer vision show that these methods achieve the required robustness whilst outperforming other precise and imprecise methods.

   

   

  Website 
• Cuzzolin F, 'On the fiber bundle structure of the space of belief functions'
  Annals of Combinatorics 18 (2) (2014) pp.245-263
  ISSN: 0218-0006 eISSN: 0219-3094
  Abstract

   

  The study of finite non-additive measures or “belief functions” has been recently posed in connection with combinatorics and convex geometry. As a matter of fact, as belief functions are completely specified by the associated belief values on the events of the frame on which they are defined, they can be represented as points of a Cartesian space. The space of all belief functions B or “belief space” is a simplex whose vertices are BF focused on single events. In this paper, we present an alternative description of the space of belief functions in terms of differential geometric notions. The belief space possesses indeed a recursive bundle structure inherently related to the mass assignment mechanism, in which basic probability is recursively assigned to events of increasing size. A formal proof of the decomposition of B together with a characterization of its bases and fibers as simplices are provided.

  Website 
• Cuzzolin F, Mateus D, Horaud R, 'Robust temporally coherent laplacian protrusion segmentation of 3D articulated bodies'
  International Journal of Computer Vision 112 (1) (2014) pp.43-70
  ISSN: 0920-5691
  Abstract

  In motion analysis and understanding it is important to be able to fit a suitable model or structure to the temporal series of observed data, in order to describe motion patterns in a compact way, and to discriminate between them. In an unsupervised context, i.e., no prior model of the moving object(s) is available, such a structure has to be learned from the data in a bottom-up fashion. In recent times, volumetric approaches in which the motion is captured from a number of cameras and a voxel-set representation of the body is built from the camera views, have gained ground due to attractive features such as inherent view-invariance and robustness to occlusions. Automatic, unsupervised segmentation of moving bodies along entire sequences, in a temporally-coherent and robust way, has the potential to provide a means of constructing a bottom-up model of the moving body, and track motion cues that may be later exploited for motion classification. Spectral methods such as locally linear embedding can be useful in this context, as they preserve “protrusions”, i.e., high-curvature regions of the 3D volume, of articulated shapes, while improving their separation in a lower dimensional space, making them in this way easier to cluster. In this paper we therefore propose a spectral approach to unsupervised and temporally-coherent body-protrusion segmentation along time sequences. Volumetric shapes are clustered in an embedding space, clusters are propagated in time to ensure coherence, and merged or split to accommodate changes in the body’s topology. Experiments on both synthetic and real sequences of dense voxel-set data are shown. This supports the ability of the proposed method to cluster body-parts consistently over time in a totally unsupervised fashion, its robustness to sampling density and shape quality, and its potential for bottom-up model construction.

   

  Website 
• Sapienza M, Cuzzolin F, Torr PHS, 'Learning discriminative space-time action parts from weakly labelled videos'
  International Journal of Computer Vision 110 (1) (2014) pp.30-47
  ISSN: 0920-5691
  Abstract

  Current state-of-the-art action classification methods aggregate space–time features globally, from the entire video clip under consideration. However, the features extracted may in part be due to irrelevant scene context, or movements shared amongst multiple action classes. This motivates learning with local discriminative parts, which can help localise which parts of the video are significant. Exploiting spatio-temporal structure in the video should also improve results, just as deformable part models have proven highly successful in object recognition. However, whereas objects have clear boundaries which means we can easily define a ground truth for initialisation, 3D space–time actions are inherently ambiguous and expensive to annotate in large datasets. Thus, it is desirable to adapt pictorial star models to action datasets without location annotation, and to features invariant to changes in pose such as bag-of-feature and Fisher vectors, rather than low-level HoG. Thus, we propose local deformable spatial bag-of-features in which local discriminative regions are split into a fixed grid of parts that are allowed to deform in both space and time at test-time. In our experimental evaluation we demonstrate that by using local space–time action parts in a weakly supervised setting, we are able to achieve state-of-the-art classification performance, whilst being able to localise actions even in the most challenging video datasets.

  Website 
• Cuzzolin F, Sapienza M, 'Learning Pullback HMM Distances'
  IEEE Transactions on Pattern Analysis and Machine Intelligence 36 (7) (2014) pp.1483-1489
  ISSN: 0162-8828
  Abstract
  Recent work in action recognition has exposed the limitations of methods which directly classify local features extracted from spatio-temporal video volumes. In opposition, encoding the actions' dynamics via generative dynamical models has a number of attractive features: however, using all-purpose distances for their classification does not necessarily deliver good results. We propose a general framework for learning distance functions for generative dynamical models, given a training set of labelled videos. The optimal distance function is selected among a family of pullback ones, induced by a parametrised automorphism of the space of models. We focus here on hidden Markov models and their model space, and design an appropriate automorphism there. Experimental results are presented which show how pullback learning greatly improves action recognition performances with respect to base distances.
  Website 
• Cuzzolin F, 'On the relative belief transform'
  International Journal of Approximate Reasoning 53 (5) (2012) pp.786-804
  ISSN: 0888-613X
  Abstract In this paper we discuss the semantics and properties of the relative belief transform, a probability transformation of belief functions closely related to the classical plausibility transform. We discuss its rationale in both the probability-bound and Shafer" s interpretations of belief functions. Even though the resulting probability (as it is the case for the plausibility transform) is not consistent with the original belief function, an interesting rationale in terms of optimal strategies in a non-cooperative game can be given in the probability-bound interpretation to both relative belief and plausibility of singletons. On the other hand, we prove that relative belief commutes with Dempster" s orthogonal sum, meets a number of properties which are the duals of those met by the relative plausibility of singletons, and commutes with convex closure in a similar way to Dempster" s rule. This supports the argument that relative plausibility and belief transform are indeed naturally associated with the D-S framework, and highlights a classification of probability transformations into two families, according to the operator they relate to. Finally, we point out that relative belief is only a member of a class of -œrelative mass- mappings, which can be interpreted as low-cost proxies for both plausibility and pignistic transforms.Website 
• Ozonoff A, Cuzzolin F, Snow P, 'Special issue on information fusion applications to human health and safety'
  Information Fusion 13 (2) (2012) pp.102-104
  ISSN: 1566-2535
  Abstract

  This special issue presents articles submitted in response to calls for papers in the domains of public health and of comparisons of fusion methods in real-world applications. The fruits of the two searches were combined for presentation under the new rubric, in one issue. We aim to increase awareness of and involvement in the discipline of information fusion among researchers who are unfamiliar with the discipline and might benefit from new approaches. We also hope to refocus the attention of those with experience and expertise in information fusion on new application areas, in the hopes that new partnerships and collaborations might result across disciplines. The research presented in this special issue samples the range of applied problems in human health and safety that have already seen benefits from an information fusion approach. The studies presented here offer a glimpse of large expanses of applied scientific territory still waiting to be explored. In welcoming people who might be newly aware of the field, it may be appropriate to begin with a definition.

  Website 
• Cuzzolin F, 'Credal semantics of Bayesian approximations in terms of probability intervals'
  IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics 40 (2) (2010) pp.421-432
  ISSN: 1083-4419
  Abstract In this paper we propose a credal representation of the interval probability associated with a belief function (b.f.), and show how it relates to several classical Bayesian transformations of b.f.s through the notion of -œfocus- of a pair of simplices. While a belief function corresponds to a polytope of probabilities consistent with it, the related interval probability is geometrically represented by a pair of upper and lower simplices. Starting from the interpretation of the pignistic function as the center of mass of the credal set of consistent probabilities, we prove that relative belief of singletons, relative plausibility of singletons and intersection probability can all be described as foci of different pairs of simplices in the region of all probability measures. The formulation of frameworks similar to the Transferable Belief Model for such Bayesian transformations appears then at hand.Website 
• Cuzzolin F, 'Geometry of relative plausibility and relative belief of singletons'
  Annals of Mathematics and Artificial Intelligence 59 (1) (2010) pp.47-79
  ISSN: 1012-2443
  Abstract The study of the interplay between belief and probability has recently been posed in a geometric framework, in which belief and plausibility functions are represented as points of simplices in a Cartesian space. All Bayesian approximations of a belief function b form two homogeneous groups, which we call "affine" and "epistemic" families. In this paper, in particular, we focus on relative plausibility and belief of singletons and show that they form, together with a new Bayesian function called "non-Bayesianity flag", a homogenous family of Bayesian functions related to b, in terms of both their geometry and their behavior with respect to Dempster's rule of combination. We investigate their geometry, which turns out to be described in terms of three planes and angles. We combine algebraic and geometric properties of the relative plausibility function to conjecture its new interpretation as solution of the probabilistic approximation problem formulated in terms of the rule of combination. Finally, drawing inspiration from the binary case, we prove that all Bayesian approximations of both families coincide when belief functions assign the same mass to events of the same size.Website 
• Cuzzolin F, 'The geometry of consonant belief functions: simplicial complexes of necessity measures'
  Fuzzy Sets and Systems 161 (10) (2010) pp.1459-1479
  ISSN: 0165-0114
  Abstract In this paper we extend the geometric approach to the theory of evidence in order to include the class of necessity measures, represented on a finite domain of "frame" by consonant belief functions (b.f.s). The correspondence between chains of subsets and convex sets of b.f.s is studied and its properties analyzed, eventually yielding an elegant representation of the region of consonant belief functions in terms of the notion of -œsimplicial complex-. In particular we focus on the set of outer consonant approximations of a belief function, showing that for each maximal chain of subsets these approximations form a polytope. The maximal such approximation with respect to the weak inclusion relation between b.f.s is one of the vertices of this polytope, and is generated by a permutation of the elements of the frame.Website 
• Cuzzolin F, 'Three alternative combinatorial formulations of the theory of evidence'
  Intelligent Data Analysis 14 (4) (2010) pp.439-464
  ISSN: 1088-467X
  Abstract In this paper we introduce three alternative combinatorial formulations of the theory of evidence (ToE), by proving that both plausibility and commonality functions share the structure of "sum function" with belief functions. We compute their Moebius inverses, which we call basic plausibility and commonality assignments. In the framework of the geometric approach to uncertainty measures the equivalence of the associated formulations of the ToE is mirrored by the geometric congruence of the related simplices. We can then describe the point-wise geometry of these sum functions in terms of rigid transformations mapping them onto each other. Combination rules can be applied to plausibility and commonality functions through their Moebius inverses, leading to interesting applications of such inverses to the probabilistic transformation problem.Website 
• Cuzzolin F, 'A geometric approach to the theory of evidence'
  IEEE Transactions on Systems, Man, and Cybernetics, Part C: Applications and Reviews 38 (4) (2008) pp.522-534
  ISSN: 1094-6977
  Abstract In this paper, we propose a geometric approach to the theory of evidence based on convex geometric interpretations of its two key notions of belief function (b.f.) and Dempster's sum. On one side, we analyze the geometry of b.f.'s as points of a polytope in the Cartesian space called belief space, and discuss the intimate relationship between basic probability assignment and convex combination. On the other side, we study the global geometry of Dempster's rule by describing its action on those convex combinations. By proving that Dempster's sum and convex closure commute, we are able to depict the geometric structure of conditional subspaces, i.e., sets of b.f.'s conditioned by a given function b. Natural applications of these geometric methods to classical problems such as probabilistic approximation and canonical decomposition are outlined.Website 

Book chapters

• Saha S, Singh G, Sapienza M, Torr P, Cuzzolin F, 'Spatio-Temporal Action Instance Segmentation and Localisation' in Noceti N, Sciutti A, Rea F (ed.), Modelling Human Motion, Springer (2020)
  ISBN: 9783030467319 eISBN: 9783030467326
  Abstract

  Current state-of-the-art human action recognition is focused on the classification of temporally trimmed videos in which only one action occurs per frame. In this work we address the problem of action localisation and instance segmentation in which multiple concurrent actions of the same class may be segmented out of an image sequence. We cast the action tube extraction as an energy maximisation problem in which configurations of region proposals in each frame are assigned a cost and the best action tubes are selected via two passes of dynamic programming. One pass associates region proposals in space and time for each action category, and another pass is used to solve for the tube’s temporal extent and to enforce a smooth label sequence through the video. In addition, by taking advantage of recent work on action foreground-background segmentation, we are able to associate each tube with class-specific segmentations. We demonstrate the performance of our algorithm on the challenging LIRIS-HARL dataset and achieve a new state-of-the-art result which is 14.3 times better than previous methods.

  Website 
• Cuzzolin F, 'An algebraic study of the notion of independence of frames' in Chakraverty S (ed.), Mathematics of Uncertainty Modeling in the Analysis of Engineering and Science Problems, IGI Publishing (2014)
  ISBN: 9781466649910 eISBN: 9781466649927
  Abstract The theory of belief functions or “theory of evidence” allows the mathematical representation of uncertain pieces of evidence on which decisions can be based. Frequently, different pieces of evidence belong to distinct, albeit related, domains or “frames”: for instance, audio and video clues can be combined to infer the identity of a person from a video. Evidence encoded by different belief functions on separate frames can be merged on a common frame, a combination which is guaranteed to exist if and only if the frames are “independent” in the sense of Boolean algebras. In all other cases the evidence conflicts. Independence of frames and belief function combinability are then strictly related. In this chapter, the authors discuss the notion of independence of frames in the theory of evidence from an algebraic point of view, starting from an analogy with standard linear independence. The final goal is to search for a solution of the problem of conflicting belief function via a generalization of the classical Gram-Schmidt algorithm for vector orthogonalization. Families of frames can be given several algebraic interpretations in terms of semi-modular lattices, matroids, and geometric lattices. Each of those structures is endowed with a particular (extended) independence relation, which we prove to be distinct albeit related to independence of frames.Website 
• Cuzzolin F, 'Complexes of outer consonant approximations' in Sossai C, Chemello G (ed.), Symbolic and quantitative approaches to reasoning with uncertainty:Lecture notes in computer science, vol 5590, Springer Berlin / Heidelberg (2009)
  ISBN: 9783642029059
  Abstract In this paper we discuss the problem of approximating a belief function (b.f.) with a necessity measure or “consonant belief function” (co.b.f.) from a geometric point of view. We focus in particular on outer consonant approximations, i.e. co.b.f.s less committed than the original b.f. in terms of degrees of belief. We show that for each maximal chain of focal elements the set of outer consonant approximation is a polytope. We describe the vertices of such polytope, and characterize the geometry of maximal outer approximations.Website 
• Cuzzolin F, 'The Intersection Probability and Its Properties' in Sossai C, Chemello G (ed.), Symbolic and Quantitative Approaches to Reasoning with Uncertainty, Springer Berlin Heidelberg (2009)
  ISBN: 978-3-642-02905-9 eISBN: 978-3-642-02906-6
  Abstract In this paper we discuss the properties of the intersection probability, a recent Bayesian approximation of belief functions introduced by geometric means. We propose a rationale for this approximation valid for interval probabilities, study its geometry in the probability simplex with respect to the polytope of consistent probabilities, and discuss the way it relates to important operators acting on belief functions.Website 
• Cuzzolin F, 'Alternative Formulations of the Theory of Evidence Based on Basic Plausibility and Commonality Assignments' in Ho T, Zhou Z (ed.), PRICAI 2008: Trends in Artificial Intelligence, Springer Berlin Heidelberg (2008)
  ISBN: 978-3-540-89196-3
  Abstract In this paper we introduce indeed two alternative formulations of the theory of evidence by proving that both plausibility and commonality functions share the same combinatorial structure of sum function of belief functions, and computing their Moebius inverses called basic plausibility and commonality assignments. The equivalence of the associated formulations of the ToE is mirrored by the geometric congruence of the related simplices. Applications to the probabilistic approximation problem are briefly presented.Website 
• Cuzzolin F, 'Dual Properties of the Relative Belief of Singletons' in PRICAI 2008 Trends in Artificial Intelligence, Springer (2008)
  ISBN: 9783540891963
  Abstract

  In this paper we prove that a recent Bayesian approximation of belief functions, the relative belief of singletons, meets a number of properties with respect to Dempster’s rule of combination which mirrors those satisfied by the relative plausibility of singletons. In particular, its operator commutes with Dempster’s sum of plausibility functions, while perfectly representing a plausibility function when combined through Dempster’s rule. This suggests a classification of all Bayesian approximations into two families according to the operator they relate to.

   

  Website 

Conference papers

• Hossain M, Cannon K, Jang D, Cuzzolin F, Xu Z, 'Video-based crowd counting using a multi-scale optical flow pyramid network'
  (2021) pp.3-20
  ISSN: 0302-9743
  Abstract

  This paper presents a novel approach to the task of videobased crowd counting, which can be formalized as the regression problem of learning a mapping from an input image to an output crowd density map. Convolutional neural networks (CNNs) have demonstrated striking accuracy gains in a range of computer vision tasks, including crowd counting. However, the dominant focus within the crowd counting literature has been on the single-frame case or applying CNNs to videos in a frame-by-frame fashion without leveraging motion information. This paper proposes a novel architecture that exploits the spatiotemporal information captured in a video stream by combining an optical flow pyramid with an appearance-based CNN. Extensive empirical evaluation on five public datasets comparing against numerous state-of-the-art approaches demonstrates the efficacy of the proposed architecture, with our methods reporting best results on all datasets. Finally, a set of transfer learning experiments shows that, once the proposed model is trained
  on one dataset, it can be transferred to another using a limited number of training examples and still exhibit high accuracy.
  

  Website 
• Gune O, Banerjee B, Chaudhuri S, Cuzzolin F, 'Generalized zero-shot learning using generated proxy unseen samples and entropy separation'
  (2020) pp.4262-4270
  ISBN: 9781450379885
  Abstract

  The recent generative model-driven Generalized Zero-shot Learning (GZSL) techniques overcome the prevailing issue of the model bias towards the seen classes by synthesizing the visual samples of the unseen classes through leveraging the corresponding semantic prototypes. Although such approaches significantly improve the GZSL performance due to data augmentation, they violate the principal assumption of GZSL regarding the unavailability of semantic information of unseen classes during training. In this work, we propose to use a generative model (GAN) for synthesizing the visual proxy samples while strictly adhering to the standard assumptions of the GZSL. The aforementioned proxy samples are generated by exploring the early training regime of the GAN. We hypothesize that such proxy samples can effectively be used to characterize the average entropy of the label distribution of the samples from the unseen classes. Further, we train a classifier on the visual samples from the seen classes and proxy samples using entropy separation criterion such that an average entropy of the label distribution is low and high, respectively, for the visual samples from the seen classes and the proxy samples. Such entropy separation criterion generalizes well during testing where the samples from the unseen classes exhibit higher entropy than the entropy of the samples from the seen classes. Subsequently, low and high entropy samples are classified using supervised learning and ZSL rather than GZSL. We show the superiority of the proposed method by experimenting on AWA1, CUB, HMDB51, and UCF101 datasets.
  

  Website 
• Zhu H, Liu D, Bayley I, Harrison R, Cuzzolin F, 'Datamorphic Testing: A Method for Testing Intelligent Applications'
  (2019) pp.149-156
  ISBN: 9781728104935 eISBN: 9781728104928
  Abstract

  Adequate testing of AI applications is essential to ensure their quality. However, it is often prohibitively difficult to generate realistic test cases or to check software correctness. This paper proposes a new method called datamorphic testing, which consists of three components: a set of seed test cases, a set of datamorphisms for transforming test cases, and a set of metamorphisms for checking test results. With an example of face recognition application, the paper demonstrates how to develop datamorphic test frameworks, and illustrates how to perform testing in various strategies, and validates the approach using an experiment with four real industrial applications of face recognition.

  Website 
• Singh G, Saha S, Cuzzolin F, 'Predicting action tubes'
  (2019) pp.106-123
  ISSN: 0302-9743 ISBN: 9783030110147
  Abstract

  In this work, we present a method to predict an entire ‘action tube’ (a set of temporally linked bounding boxes) in a trimmed video just by observing a smaller subset of it. Predicting where an action is going to take place in the near future is essential to many computer vision based applications such as autonomous driving or surgical robotics. Importantly, it has to be done in realtime and in an online fashion. We propose a Tube Prediction network (TPnet) which jointly predicts the past, present and future bounding boxes along with their action classification scores. At test time TPnet is used in a (temporal) sliding window setting, and its predictions are put into a tube estimation framework to construct/predict the video long action tubes not only for the observed part of the video but also for the unobserved part. Additionally, the proposed action tube predictor helps in completing action tubes for unobserved segments of the video. We quantitatively demonstrate the latter ability, and the fact that TPnet improves state-of-the-art detection performance, on one of the standard action detection benchmarks - J-HMDB-21 dataset.

  
  
  Website 
• Cuzzolin F, 'Generalised max entropy classifiers'
  (2018) pp.39-47
  ISSN: 0302-9743 eISSN: 1611-3349 ISBN: 9783319993829
  Abstract

  In this paper we propose a generalised maximum-entropy classification framework, in which the empirical expectation of the feature functions is bounded by the lower and upper expectations associated with the lower and upper probabilities associated with a belief measure. This generalised setting permits a more cautious appreciation of the information content of a training set. We analytically derive the KarushKuhn-Tucker conditions for the generalised max-entropy classifier in the case in which a Shannon-like entropy is adopted.

  Website 
• Cuzzolin F, 'General geometry of belief function combination'
  (2018) pp.48-56
  ISSN: 0302-9743 eISSN: 1611-3349 ISBN: 9783319993829
  Abstract

  In this paper we build on previous work on the geometry of Dempster’s rule to investigate the geometric behaviour of various other combination rules, including Yager’s, Dubois’, and disjunctive combination, starting from the case of binary frames of discernment. Believability measures for unnormalised belief functions are also considered. A research programme to complete this analysis is outlined.

  Website 
• Behl H, Sapienza M, Singh G, Saha S, Cuzzolin F, Torr P, 'Incremental Tube Construction for Human Action Detection'
  (2018)
  Abstract

  Current state-of-the-art action detection systems are tailored for offline batch-processing applications. However, for online applications like human-robot interaction, current systems fall short. In this work, we introduce a real-time and online joint-labelling and association algorithm for action detection that can incrementally construct space-time action tubes on the most challenging untrimmed action videos in which different action categories occur concurrently. In contrast to previous methods, we solve the linking, action labelling and temporal localization problems jointly in a single pass. We demonstrate superior online association accuracy and speed (1.8ms per frame) as compared to the current state-of-the-art offline and online systems.
  

  Website 
• Saha S, Singh G, Cuzzolin F, 'AMTnet: Action-Micro-Tube regression by end-to-end trainable deep architecture'
  (2017) pp.4414-4423
  ISSN: 2380-7504 ISBN: 9781538610329
  Abstract

  Dominant approaches to action detection can only provide sub-optimal solutions to the problem, as they rely on seeking frame-level detections, to later compose them into ‘action tubes’ in a post-processing step. With this paper we radically depart from current practice, and take a first step towards the design and implementation of a deep network architecture able to classify and regress whole video subsets, so providing a truly optimal solution of the action detection problem. In this work, in particular, we propose a novel deep net framework able to regress and classify 3D region proposals spanning two successive video frames, whose core is an evolution of classical region proposal networks (RPNs). As such, our 3D-RPN net is able to effectively encode the temporal aspect of actions by purely exploiting appearance, as opposed to methods which heavily rely on expensive flow maps. The proposed model is end-to-end trainable and can be jointly optimised for action localisation and classification in a single step. At test time the network predicts ‘micro-tubes’ encompassing two successive frames, which are linked up into complete action tubes via a new algorithm which exploits the temporal encoding learned by the network and cuts computation time by 50%. Promising results on the J-HMDB-21 and UCF-101 action detection datasets show that our model does outperform the state-of-the-art when relying purely on appearance.

  Website 
• Singh G, Saha S, Sapienza M, Torr P, Cuzzolin F, 'Online Real-time Multiple Spatiotemporal Action Localisation and Prediction'
  (2017) pp.3637-3646
  ISSN: 2380-7504 ISBN: 9781538610329
  Abstract

  We present a deep-learning framework for real-time multiple spatio-temporal (S/T) action localisation, classification and early prediction. Current state-of-the-art approaches work offline, and are too slow be useful in realworld settings. To overcome their limitations we introduce two major developments. Firstly, we adopt real-time SSD (Single Shot MultiBox Detector) convolutional neural networks to regress and classify detection boxes in each video frame potentially containing an action of interest. Secondly, we design an original and efficient online algorithm to incrementally construct and label ‘action tubes’ from the SSD frame level detections. As a result, our system is not only capable of performing S/T detection in real time, but can also perform early action prediction in an online fashion. We achieve new state-of-the-art results in both S/T action localisation and early action prediction on the challenging UCF101-24 and J-HMDB-21 benchmarks, even when compared to the top offline competitors. To the best of our knowledge, ours is the first real-time (up to 40fps) system able to perform online S/T action localisation and early action prediction on the untrimmed videos of UCF101-24.
  

  Website 
• Zhou C, Cuzzolin F, 'The total belief theorem'
  (2017)
  Abstract

  In this paper, motivated by the treatment of conditional constraints in the data association problem, we state and prove the generalisation of the law of total probability to belief functions, as finite random sets. Our results apply to the case in which Dempster’s conditioning is employed. We show that the solution to the resulting total belief problem is in general not unique, whereas it is unique when the a-priori belief function is Bayesian. Examples and case studies underpin the theoretical contributions. Finally, our results are compared to previous related work on the generalisation of Jeffrey’s rule by Spies and Smets.
  

  Website 
• Roman C, Sapienza M, Ball P, Ou S, Cuzzolin F, Torr PHS, 'Heterogeneous Wireless System Testbed for Remote Image Processing in Automated Vehicles'
  (2016)
  ISBN: 978-1-5090-2526-8
  Abstract Automated vehicles will carry computing and communication platforms, and will have enhanced sensing capabilities. Safety around people along with obstacle detection and avoidance systems are key to their success. In controlled environments, automated vehicles can benefit from a remote processing approach to reduce cost and accelerate deployment on larger scales. In this paper we present a section of our intelligent transport systems testbed which evaluates the remote image processing approach with a novel heterogeneous wireless communication system. Hardware implementation is carried out for an experimental evaluation and comparison with the simulation results.Website 
• Saha S, Singh G, Sapienza M, Torr P, Cuzzolin F, 'Deep Learning for Detecting Multiple Space-Time Action Tubes in Videos'
  (2016)
  Abstract

  In this work, we propose an approach to the spatiotemporal localisation (detection) and classification of multiple concurrent actions within temporally untrimmed videos. Our framework is composed of three stages. In stage 1, appearance and motion detection networks are employed to localise and score actions from colour images and optical flow. In stage 2, the appearance network detections are boosted by combining them with the motion detection scores, in proportion to their respective spatial overlap. In stage 3, sequences of detection boxes most likely to be associated with a single action instance, called action tubes, are constructed by solving two energy maximisation problems via dynamic programming. While in the first pass, action paths spanning the whole video are built by linking detection boxes over time using their class-specific scores and their spatial overlap, in the second pass, temporal trimming is performed by ensuring label consistency for all constituting detection boxes. We demonstrate the performance of our algorithm on the challenging UCF101, J-HMDB-21 and LIRIS-HARL datasets, achieving new state-of-the-art results across the board and significantly increasing detection speed at test time.

  Website 
• Jetley S, Cuzzolin F, '3D activity recognition using gradient analysis consolidated over motion history and binary shape templates'
  (2014) pp.129-144
  ISSN: 0302-9743
  Abstract

  This paper presents our work on activity recognition in 3D depth images. We propose a global descriptor that is accurate, compact and easy to compute as compared to the state-of-the-art for characterizing depth sequences. Activity enactment video is divided into temporally overlapping blocks. Each block (set of image frames) is used to generate Motion History Templates (MHTs) and Binary Shape Templates (BSTs) over three different views - front, side and top. The three views are obtained by projecting each video frame onto three mutually orthogonal Cartesian planes. MHTs are assembled by stacking the difference of consecutive frame projections in a weighted manner separately for each view. Histograms of oriented gradients are computed and concatenated to represent the motion content. Shape information is obtained through a similar gradient analysis over BSTs. These templates are built by overlaying all the body silhouettes in a block, separately for each view. To effectively trace shape-growth, BSTs are built additively along the blocks. Consequently, the complete ensemble of gradient features carries both 3D shape and motion information to effectively model the dynamics of an articulated body movement. Experimental results on 4 standard depth databases (MSR 3D Hand Gesture, MSR Action, Action-Pairs, and UT-Kinect) prove the efficacy as well as the generality of our compact descriptor. Further, we successfully demonstrate the robustness of our approach to (impulsive) noise and occlusion errors that commonly affect depth data.

  Website 
• De Rosa R, Cesa-Bianchi N, Gori I, Cuzzolin F, 'Online action recognition via nonparametric incremental learning'
  (2014) pp.1-15
  Abstract

  We introduce an online action recognition system that can be combined with any set of frame-by-frame feature descriptors. Our system covers the frame feature space with classifiers whose distribution adapts to the hardness of locally approximating the Bayes optimal classifier. An efficient nearest neighbour search is used to find and combine the local classifiers that are closest to the frames of a new video to be classified. The advantages of our approach are: incremental training, frame by frame real-time prediction, nonparametric predictive modelling, video segmentation for continuous action recognition, no need to trim videos to equal lengths and only one tuning parameter (which, for large datasets, can be safely set to the diameter of the feature space). Experiments on standard benchmarks show that our system is competitive with state-of-the-art non-incremental and incremental baselines. keywords: action recognition, incremental learning, continuous action recognition, nonparametric model, real time, multivariate time series classification, temporal classification

  Website 
• Cuzzolin F, 'Generalizations of the Relative Belief Transform'
  164 (2013) pp.109-116
  Abstract
  Probability transformation of belief functions can be classified into different
  families, according to the operator they commute with. In particular, as they
  commute with Dempster’s rule, relative plausibility and belief transforms form one
  such “epistemic” family, and possess natural rationales within Shafer’s formulation
  of the theory of evidence. However, the relative belief transform only exists when
  some mass is assigned to singletons. We show here that relative belief is only a
  member of a class of “relative mass” mappings, which can be interpreted as lowcost
  proxies for both plausibility and pignistic transforms.
  Website 
• Cuzzolin F, 'Game-theoretical Semantics of Epistemic Probability Transformations'
  164 (2013) pp.101-108
  Abstract

  Probability transformation of belief functions can be classified into different families, according to the operator they commute with. In particular, as they commute with Dempster’s rule, relative plausibility and belief transforms form one such “epistemic” family, and possess natural rationales within Shafer’s formulation of the theory of evidence, while they are not consistent with the credal or probability-bound semantic of belief functions. We prove here, however, that these transforms can be given in this latter case an interesting rationale in terms of optimal strategies in a non-cooperative game.

  Website 
• Cuzzolin F, 'L-p Consonant Approximation of Belief Functions in the Mass Space'
  (2013) pp.149-158
  Abstract
  Abstract—In this paper, we solve the problem of approximating a belief measure with a necessity measure or “consonant belief function” in a geometric framework. Consonant belief functions form a simplicial complex in both the space of all belief functions and the space of all mass vectors: Partial approximations are first sought in each component of the complex, while global solutions are selected among them. As a first step in this line of study, we seek
  here approximations that minimize Lp norms. Approximations in the mass space can be interpreted in terms of mass redistribution, while approximations in the belief space generalize the maximal outer consonant approximation.We compare them with each other and with other classical approximations and illustrate them with the help of a running example.
• Cuzzolin F, 'Consonant Approximations in the Belief Space'
  164 (2013) pp.125-133
  Abstract
  In this paper we solve the problem of approximating a belief measure with a necessity measure or “consonant belief function” by minimizing appropriate distances from the consonant complex in the space of all belief functions. Partial approximations are first sought in each simplicial component of the consonant complex, while global solutions are obtained from the set of partial ones. The L1, L2 and L¥ consonant approximations in the belief space are here computed, discussed and interpreted as generalizations of the maximal outer consonant approximation.
  Results are also compared to other classical approximations in a ternary example.
  Website 
• Wenjuan Gong, Michael Sapienza, and Fabio Cuzzolin, 'Fisher Tensor Decomposition for Unconstrained Gait Recognition'
  (2013)
  Website 
• Cuzzolin F, Wenjuan Gong, 'Belief modeling regression for pose estimation'
  (2013) pp.1398-1405
  ISSN: 978-605-86311-1-3
  Website 
• Cuzzolin F, 'Credal Semantics of Bayesian Transformations'
  (2010) pp.129-138
  
• Cuzzolin F, 'Consistent approximations of belief functions'
  (2009) pp.139-148
  Abstract Consistent belief functions represent collections of coherent or non-contradictory pieces of evidence. As most operators used to update or elicit evidence do not preserve consistency, the use of consistent transformations cs[.] in a reasoning process to guarantee coherence can be desirable. Such transformations are turn linked to the problem of approximating an arbitrary belief function with a consistent one. We study here the consistent approximation problem in the case in which distances are measured using classical Lp norms. We show that, for each choice of the element we want them to focus on, the partial approximations determined by the L1 and L2 norms coincide, and can be interpreted as classical focused consistent transformations. Global L1 and L2 solutions do not in general coincide, however, nor are they associated with the highest plausibility element.
• Cuzzolin F, 'Learning pullback metrics for linear models'
  (2008)
  Abstract In this paper we present an unsupervised differential-geometric approach for learning Riemannian metrics for dynamical models. Given a training set of models the optimal metric is selected among a family of pullback metrics induced by the Fisher information tensor through a parameterized diffeomorphism. The problem of classifying motions, en- coded as dynamical models of a certain class, can then be posed on the learnt manifold. Experimental results concerning action and iden- tity recognition based on simple scalar features are shown, proving how learning a metric actually improves classifcation rates when compared with Fisher geodesic distance and other classical distance functions.Website 
• Cuzzolin F, 'On the Credal Structure of Consistent Probabilities'
  5293 (2008) pp.126-139
  

Books

• Cuzzolin, The geometry of uncertainty – The geometry of imprecise probabilities, Artificial Intelligence: Foundations, Theory and Algorithms, Springer (to appear in 2018)
• Cuzzolin, S. Destercke, T. Denoeux, A. Martin (Editors), Belief Functions: Theory and Applications - Proceedings of the 5th International Conference on Belief Functions, Springer, 2018 (upcoming)

Journals

• Mellema and F. Cuzzolin, ISIF Sponsored Workshop - BELIEF 2014, ISIF Perspectives on Information Fusion, 2016
• F. Cuzzolin, Two new Bayesian approximations of belief functions based on convex geometry, IEEE Tr. Systems, Man, and Cybernetics - part B 37(4), pp. 993-1008, 2007 (IF 7.384)
• Cuzzolin, Algebraic structure of the families of compatible frames of discernment, Ann Math Artif Intel 45, 241-274, 2005
• F. Cuzzolin, Geometry of Dempster's rule of combination, IEEE Tr. SMC - part B 34(2), 961- 977, 2004 (IF 7.384)

Areas of expertise

Artificial Intelligence, Uncertainty Theory, Belief Functions, Machine Learning, Computer vision, Robotics.

Membership of professional bodies

Editorships

I was Associate Editor of the IEEE Transactions on Fuzzy Systems (September 2013 - 2017), the top journal in Computer Science with a 2015 impact factor of 8.746. I was also Associate Editor of the IEEE Transactions on Systems, Man, and Cybernetics C (January 2011 - March 2013), Guest Editor of Elsevier’s International Journal of Approximate Reasoning (October 2014 - present), Guest Editor of Elsevier’s Information Fusion (2009-12).

I am currently Associate Editor for the International Journal of Approximate Reasoning (ICAR).

Grant panel memberships

I am a Reviewer for the European Research Council (ERC), Consolidator Grants (CoG), the Italian Ministry of Scientific Research, and I am Member of the Associate College of reviewers for EPSRC.

Invited talks and keynotes

I have been invited to give seminars in a host of national and international institutions, including Oxford University’s Department of Engineering (Feb 13 2018), Harvard University’s Department of Statistics (July 14 2016), Cambridge University’s Department of Engineering (March 2016), Microsoft Research Europe (2006), GeorgiaTech (2006), Pompeu Fabra University (2006), EPFL-IDIAP (2006), MIT (1999).

In 2015 I was invited by the Oxford Martin School to join a select group of Oxford Academics to meet with former World Chess Champion Garry Kasparov and discuss some of the future challenges of AI. I was invited again in March 2017.

In the last two years I was invited speaker orkeynote speaker at a number of events:

• the CSA 2016 international conference (Algiers, Dec 2016);
• the 2017 BMW Knowledge Day (February 2017);
• the 4^th Summer School on Belief Functions (Xian, China, July 2017);
• the Roadmap to Autonomous Surgery Workshop, Verona, Italy, Oct 31 2017;
• the Oxford Prospects Programme (St Cross and Regent’s colleges, Oxford, Jan 31/Jul 31 2018);
• the Ambassadors’ Roundtable on AI of the Anglo-Israel Association (Royal Society, London, Feb 27 2018);
• the Bayesian, Fiducial and Frequentist Workshop (BFF4), Harvard, May 2017 (refused for family reasons);
• the Fifth Bayesian, Fiducial and Frequentist Workshop (BFF5), Ann Arbor, USA, May 6-9 2018;
• the 2^nd Biennial Summer School on Surgical Robotics (COSUR 2018), Verona, Italy, July 10 2018;
• ICRA 2019 workshop: “Next Generation Robotic Surgery: Seamless integration of Machine Learning, Knowledge Representation and Robotics within the operating rooms”, Montreal, Canada, May 2019 https://nxgsur-icra2019.sciencesconf.org/;

Hamlyn Symposium Workshop: “Towards robotic autonomy in surgery”, London, June 23 2019, https://www.ukras.org/hamlyn/workshops/towards-robotic-autonomy-in-surgery/

Tutorials and Special Issues

• Belief functions for the working scientist, Half-day tutorial @ IJCAI-16
• Belief functions for the working scientist, Tutorial @Uncertainty in Artificial Intelligence (UAI) 2015
• Special Issue: Selected papers from BELIEF 2014, International Journal of Approximate Reasoning, 2015-16
• Special Issue: Information Fusion Applications to Human Health and Safety, Information Fusion, Elsevier, Volume 13, Issue 2, 2012; Co-editors: Paul Snow and Al Ozonoff (Harvard University).

Consultancy

I am Board committee member of the joint Simon Fraser University - Huawei Technologies research centre, based in Vancouver, Canada.

Conferences

Conference chairing

I was the General Chair and local organizer of the 3rd International Conference on the Theory of Belief Functions (BELIEF 2014), held in St. Hugh's college, Oxford, UK, September 26-28 2014. I was Co-organizer of WPMSIIP 2011 - 4th Int. Workshop on Principles and Methods of Statistical Inference with Interval Probability. I was in the Steering Committee (Program Chairs) of the joint SMPS-BELIEF 2018 international conference.

Conference program committees

I have served in the Technical Program Committee of almost 100 international conferences, including the top venues in Artificial Intelligence, Machine Learning, Computer Vision and Uncertainty Theory: IJCAI 2016-2019, ECAI 2018 - the top conference in Artificial Intelligence; NIPS 2018-2019 (the top machine learning venue); ICML 2018-2019 (the International Conference on Machine Learning); UAI (Uncertainty in Artificial Intelligence) 2014-2019 as Senior Program Committee member (Area Chair); BMVC (the British Machine Vision Conference)2009-2019, as Area Chair since 2016; ICCV (the International Conference on Computer Vision); IEEE CVPR (Computer Vision and Pattern Recognition); SMC (the IEEE International Conference on Systems, Man, and Cybernetics) 2013-2019, Main Track. Others major venues include FUSION 2017-19 - the International Conference on Information Fusion, IPMU 2014-2018 - the Information Processing and Management of Uncertainty series of conferences; ACCV 2014, 2018 - The Asian Conference of Computer Vision; VISAPP 2006-2018 - International Conference on Computer Vision Theory and Applications; FLAIRS 2008-2018 - the Florida AI Research Society International Conferences; ISIPTA 2009-2019 - the International Symposium on Imprecise Probabilities and Their Applications; ECSQARU 2009-19 - the Eur. Conf. on Symbolic and Quantitative Approaches to Reasoning with Uncertainty; BELIEF 2010-2018 - The International Conference on the Theory of Belief Functions; Cyberpatterns 2014 - the 3rd International Workshop on Cyberpatterns, in conjunction with SOSE2014; AVSS 2013-2014 - the IEEE International Conference on Advanced Video and Signal-Based Surveillance; IUKM2010-2019 - the Int. Symposium on Integrated Uncertainty in Knowledge Modelling and Decision Making.

Other experience

2016-now  – Professor of Artificial Intelligence, Oxford Brookes University

2008-now  – Oxford Brookes University, Department of Computing and Communication Technologies

Head of Artificial Intelligence and Vision Research Group (2012-present); Reader (September 2011-December 2015); Senior Lecturer (July 2011-September 2011); Lecturer and Early Career Fellow (September 2008-July 2011)

2006-2008 – INRIA Rhône-Alpes - Institut National de Recherche en Informatique et en Automatique: Marie Curie fellow

2004-2006 – University of California at Los Angeles, Computer Science Department: Postdoctoral researcher, UCLA Vision Lab

2003-2004 – Politecnico di Milano, Department of Electronics and Information Theory: Fixed-term Assistant Professor

2001-2003 – University of Padua, Department of Electronics & Computer Science: Postdoctoral researcher

2000          – Washington University in St Louis, Electronic Systems and Signals Research Laboratory (ESSRL): Visiting scholar

Further information

My papers have given a major contribution to the Department of Computing’s REF 2014 results. In all mock panels they have consistently scored 4* (consider that 18% outputs have eventually been judged 4*, and only 9 members of staff submitted):

• A geometric approach to the theory of evidence                       
• Geometry of relative plausibility and relative belief of singletons                       
• Credal semantics of Bayesian transformations in terms of probability intervals                        
• The geometry of consonant belief functions: simplicial complexes of necessity measures                        

Google Scholar citations: Total citations: 1474 (838 since 2014), H-index: 19, i10-index: 40.

Top papers by Scholar citation count:

1. Articulated Shape Matching Using Laplacian Eigenfunctions and Unsupervised Point Registration       (197)
2. A geometric approach to the theory of evidence                                                                            (153)
3. Deep learning for detecting multiple space-time action tubes in videos (2016)                                  (75)

Semantic Scholar citations: 1,900 citations.

Top papers by Semantic Scholar:

1. Articulated Shape Matching Using Laplacian Eigenfunctions & Unsupervised Point                         337
2. A Geometric Approach to the Theory of Evidence                                                                       219
3. Two New Bayesian Approximations of Belief Functions Based on Convex Geometry                       127
4. Deep Learning for Detecting Multiple Space-Time Action Tubes in Videos                                      90

Professor Fabio Cuzzolin was born in Jesolo, Italy.

He graduated magna cum laude with the University of Padua, was awarded a Ph.D. there in 2001 for a thesis entitled “Visions of a generalized probability theory", and worked in world-class institutions such as the Washington University in St. Louis, Politecnico di Milano, the University of California at Los Angeles. In 2006 he was awarded a Marie Curie Fellowship with INRIA Rhone-Alpes, France. In 2007 he classified second there in the Senior Researcher national recruitment.
He has been at Oxford Brookes University since 2008, took up a Senior Lectureship there in July'11, and a Readership in October 2011. He has taken on the role of Head of the Artificial Intelligence and Vision research group in September 2012, and has been awarded in October 2012 a Next 10 award by the Faculty of Technology, Design and Environment as one of its top emerging researchers.
He was promoted to Professor of Artificial Intelligence in 2016. He is currently supervising two Ph.D. students, an EPSRC-funded postdoc, and a KTP associate. Other Ph.D. students will join his group in 2016.

Professor Cuzzolin is a world expert in uncertainty theory and belief functions theory. He worked extensively on the mathematical foundations of belief calculus. His main contribution is his geometric approach to uncertainty measures, in which uncertainty measures are represented as points of a Cartesian space and there analyzed. His work on the field is in the process of being published in two separate monographs published by Springer-Verlag (“The geometry of uncertainty”) and Lambert Academic Publishing ( "Visions of a generalized probability theory").

He is also well known for his work in computer vision, mainly machine learning for human motion analysis, including tensorial models for identity recognition, metric learning for action recognition, and spectral techniques for articulated object segmentation and matching.

Further information

• Article: Walk this way, “International Innovation", a Research Media magazine, January 2014
• UAI 2015 tutorial: https://www.youtube.com/watch?v=nhGzn0R5TgM (2246 views on YouTube)
• “Towards machines that can read your mind”, Professorial Lecture, Jan 24 2018 

https://www.facebook.com/oxfordbrookes/videos/10156698398637908/ (3394 views on Facebook)

• Harvard Statistics colloquia invited talk: Belief functions: past, present and future

https://www.youtube.com/watch?v=l9XKJKgkURQ (1803 views on YouTube)

• Risk Group LLC: invited podcast on “Advances in Artificial Intelligence: Gesture and Action Recognition”

https://www.youtube.com/watch?v=xWrUwJAYqA0&t=1258s (115 views)

• “The 3,600 mile experiment: Parkinson's disease on the ocean” – MedicalXpress, June 25 2018
• “Row for Parkinson’s” – The West Australian, 7 July 2018
• “British Crew Rowing the Distance to Improve Understanding of Parkinson’s Disease”, Parkinson’s News Today, June 27 2018
• Venturefest Oxford: https://venturefestoxford.com/experts/the-crucial-role-of-theory-of-mind-tom-capabilities-in-developing-a-next-generation-human-centric-artificial-intelligence/
• BBC Inside Out East Midlands (14 January 2019): https://www.linkedin.com/feed/update/activity:6492054756503089152/