TAILOR Workpackage 9

View the Project on GitHub TAILOR-UoB/deliverables

Deliverable 9.5
(Task 9.3: PhD Training)

Mapping of AI-oriented PhD programmes at TAILOR partners

Kacper Sokol and Peter Flach, University of Bristol

Executive Summary: Understanding the availability of AI-oriented training opportunities offered by the TAILOR partners can benefit a number of tasks down the line. For example, it can be used to facilitate research visits, cross-institutional courses or even become a foundation for long-term collaboration and partnership. Universities that do not offer dedicated PhD training in AI, on the other hand, may use it as a reference material for designing bespoke PhD programmes, composing individual courses and organising thematic seminars. This document describes our approach towards mapping AI training opportunities within Europe – mostly offered by the TAILOR partners but also beyond. It lists relevant courses and materials available on the Internet and briefly annotates their content. It also touches upon different approaches to sharing the mapping insights with the TAILOR partners and the broader community. Finally, this document discusses the possibilities of designing a joint TAILOR PhD curriculum (Deliverable 9.6).


TAILOR – Trustworthy AI through Integrating Learning, Optimisation and Reasoning – is a network of over 54 partners across Europe, spanning universities and companies interested in pushing the boundaries of Artificial Intelligence (AI) research. A network of this size brings together diverse approaches and opportunities in training future generations of AI scientists, which are best exploited when properly indexed so that they are readily available to the interested parties. For example, such a resource can foster collaborations and knowledge transfer by encouraging PhD students to spend time at multiple TAILOR partners with relevant expertise. On the other hand, students looking for postgraduate courses could use it to identify suitable universities, research groups and supervisors; and companies interested in academic collaborations can scan it to find appropriate partners. Lastly, this mapping is a stepping stone towards designing a joint TAILOR PhD curriculum in AI, which can be freely adapted by the network members and beyond.


The mapping of AI-oriented training opportunities presented in this document is mainly based on Internet searches and inspection of partners’ websites. These activities were restricted to information – predominantly promotional materials about such offers – published in the English language, however the teaching opportunities were included regardless of the tuition language. The search agenda was based on the list of partners enumerated in the TAILOR network research grant proposal. Information collected in this manner was then presented to the TAILOR partners, who were asked (through a Google Form) to provide feedback, point us towards missing training opportunities and rectify any incorrect statements. These activities were followed by direct emails to leads of TAILOR partners for whom we were still missing relevant information. The inclusion criteria were built around the notion of structured PhD training opportunities in AI, such as taught courses, PhD research curricula and agendas, cohort-based PhD recruitment, and the like. The current outcome is enclosed below. We intend to maintain this as a live document so that future additions and modifications can be easily accommodated (see Dissemination Strategy, below).

Partners and Initiatives

The training opportunities identified for each partner are listed and categorised below. URLs to relevant materials are included along each partner’s name. Out of the 54 TAILOR network partners, 10 fall outside of this deliverable’s scope as they are non-educational institutions or companies.

The remaining 44 partners offer relevant PhD-level training opportunities; these are listed and indexed below. For each partner, the first sub-bullet gives a programme, followed by a list of individual modules. The partner-specific list is followed by a list of cross-institutional initiatives.

In addition to partner-specific training opportunities, cross-institutional initiatives are also common among TAILOR members. These are:

It is also worth noting the US Government Education and Training Strategy covering various AI topics delivered through The National Science Foundation’s (NSF) Computing Classroom Resources (collection of online lessons and web resources).

Additionally, various TAILOR collaborators offer relevant PhD training.

Dissemination Strategy

The mapping of AI-oriented training materials is envisaged to go through three main steps:

To streamline this process, the first and second steps for this particular deliverable take place on GitBook – a collaborative wiki/documentation online authoring application. The main reasons behind using this tool are: seamless synchronisation with GitHub repositories, document revision tracking, commenting functionality, and flexible file format underlying the GitBook documents (namely Markdown). For example, to comment on any section of this document (if it is being [re]viewed on GitBook), hover your cursor over the desired part of the document and click the + button that appears to the right; then type in your comment and save it with the up-arrow button.

Upon being accepted, this deliverable will be published as a standalone webpage through GitHub Pages. This is possible since each page of the GitBook space can be exported as a Markdown file, which is a powerful web publishing component that can be converted into many other text or markup formats (such as HTML). This means that it can also be incorporated into the TAILOR website without much overhead. Furthermore, since the mapped training opportunities can change and new offerings can become available over time, such a dissemination strategy facilitates two distinct contribution mechanisms that bypass the need for a single point of contact:

(Both approaches require a free GitHub account.)

In addition to a plain-text format, this deliverable will be published as a machine-readable JSON/YAML file linking TAILOR partners with their AI training offering annotated with relevant URLs and tags. Since the mapping results may be used in a variety of ways beyond what we originally envisaged, this alternative format will facilitate and encourage others to analyse and present the data as they see useful, e.g., with interactive visualisations. This novel dissemination strategy will increase versatility and simplify possible adoption of our findings by other European AI initiatives such as AI4media and AIDA.


This mapping is but a first step towards designing a flexible PhD curriculum and training programme in AI for TAILOR partners and beyond. The initial findings show the unexpected breadth and diversity of doctoral AI education provided in Europe. While this disparate landscape poses a major challenge, it also provides a wide selection of ideas and approaches that can be an inspiration for designing the shared curriculum. It also shows that the proposed programme needs to be flexible enough to accommodate various needs and modes of operation employed by European educational institutions, nonetheless provide a common framework for assessing and comparing skills learnt across the partners.


Towards a joint TAILOR PhD curriculum

AI curricula tend to be diverse umbrella programmes drawing from a wide range of concepts in statistics, machine learning, robotics and computer vision, among many others. To lay the foundation for a joint TAILOR PhD curriculum built from the mapped training opportunities (Deliverable 9.6), we first need to identify the main themes of AI programmes, list their auxiliary activities and discuss their organisation, coordination and realisation.

Proposed TAILOR PhD Curriculum Paradigm: Topic Tree and Badges

We propose that TAILOR should develop a Topic Forest for Artificial Intelligence Proficiency. Such a topic forest consists of a number of topic trees – see the figure below – each one describing an individual area or theme of postgraduate AI education. The tree branches will encode exposure to certain AI concepts or topics, completing each stage of which will result in attaining a badge, with the aim of reaching the terminal (leaf) node of a selected branch, which will result in a special badge.

A topic tree for arithmetic proficiency (https://www.bendevane.com/dml2015/benmiller/2015/11/02/games-learning-and-assessment-and-skill-trees/).

Expressing AI topic ontology as a collection of (possibly cross-connected) trees creates a highly flexible framework that generalises to other concepts present in AI education. For example, the branches may serve as an outline guiding educators to compose a consistent taught course. They may also help in organising a summer school or deciding on the programme of a seminar series and invited speakers (since people can also be mapped onto such trees).

We envisage the hierarchy of badges to signify competence in a well-defined area such as Data Management, AI Ethics or Bayesian Deep Learning. Based on such a collection of core topics, applied AI programmes – e.g., Medical Imaging or AI for Epidemiology – could create domain-specific units. The challenge of creating the proposed AI curriculum structure is fine-tuning the topic granularity. The guiding principle is that the topic badges should be open-ended and prescribe the minimum required training; listing additional / extracurricular topics is optional.

The badges can also describe an individual AI curriculum on a high level by visually depicting opportunities to learn about well-defined topics. Students attending such a programme would be able to select a subset of available training opportunities to steer their education in the chosen direction and complement their research. For thematic AI programmes, the students would be required to undertake training aligned with the main theme. For example, PhD programmes working with personal data ought to mandate training in data privacy, anonymisation and secure processing.

AI Training Tracks – Our Trees

Current AI training programmes tend to fall into 7 distinct categories, which will become the foundation of our forest (work in progress, Deliverable 9.6).

  1. Statistics
  2. Machine learning
  3. Data science / data engineering / big data (stages taken from DEDS)
    • Governance
    • Storage and processing
    • Preparation
    • Analysis
  4. Robotics
  5. Perception / AI for media
    • Vision (image)
    • Speech (sound)
    • Text (language)
  6. Knowledge Representation and Reasoning / Symbolic AI / Machine Reasoning
    • Logic
    • Planning
    • Reasoning
  7. Augmented Intelligence / Human–AI Interaction

Training Themes, Interdisciplinary Training and Complementary Skills

Within these tracks, an AI training programme can align with, or span across different themes. It can either be puristic; it can focus on a particular application area (e.g., healthcare); it can include interdisciplinary components (e.g., policy formation); and provide an opportunity to learn transferrable skills and develop professionally.

  1. Core AI
  2. Applied AI
    • Medical imaging
    • Healthcare
    • Medicine / medical intelligence
    • Genetics and computational biology / multiomics
    • Transport, (smart) cities
    • Finance
    • Customer relationship and support
    • Manufacturing / environment (natural disaster) / energy (management)
  3. Interdisciplinary (AI courses)
    • Humans
      • Safety
      • Philosophy and ethics
      • Fairness / accountability (/ transparency)
      • Society, policy and regulation (legal aspects)
    • Computation and Engineering
      • High-performance computing
      • Distributed and scalable (AI)
      • Distributed systems / heterogenous computing system
    • Technologies and Hardware
      • IoT
      • Blockchain
    • Miscellaneous
      • Visualisation / virtual reality / visual media / intelligent interfaces
      • Verification
  4. Transferrable Skills and Professional Development
    • Soft skills
      • Teaching and educational skills
      • Communication skills (for technical and lay audiences) – engineers, management and sales
      • Presentation skills
      • Public engagement training
    • Work
      • Career development
      • Organisational skills
      • Management skills
      • Research methods
      • Industry/partner projects; internships; placements
      • Group projects / collaboration in multi-disciplinary teams (academic and industrial settings)
    • Entrepreneurship
      • Intellectual property rights
      • Entrepreneurship / commercialisation

Organisation and Modes of Operation

Different organisation alignments underly AI training programmes.

There are also different models of operation.

This work will be continued as part of Deliverable 9.6, with the aim of building a TAILOR-branded AI PhD curriculum.