Fully Funded EPSRC PhD Studentship - The Dawn of Developmental Machine Learning: from Data Annotation to Theory Building

; Cardiff University, United Kingdom

Updated: 3 months ago
Location: Cardiff, WALES
Job Type: FullTime
Deadline: 23 Feb 2024

Studentship summary

Action is at the core of human development. Even young infants actively sample and select aspects of their environment by moving their eyes, reaching towards people and objects and eventually locomoting. Thus, young children actively construct their visual datasets for learning. But many children (especially those with neurodevelopmental conditions) experience pervasive perturbations in the motor system.

How might motor difficulties constrain the everyday visual experiences of young children?

This studentship, developed in two stages, will create a bridge between developmental science and machine learning (computer vision) to answer this question.

  • Stage 1:Machine learning as an annotation tool

Working on a unique existing dataset which combines egocentric point-of-view (head-mounted eye-tracking) with third-person views (wall-mounted cameras) during parent-child interaction in children with/without motor difficulties, the student will:

  • Characterise egocentric view properties:
  • Identify and quantify size of objects/faces/hands
  • Quantify motion blur
  • Characterise motor constraints:
  • Identify limbs and their position
  • Quantify postural stability

    • During this stage, the student will capitalise on already existing computer vision tools (e.g., OpenPose, MediaPipe), testing how accurate and precise these tools are for the current dataset. We foresee that further training of these algorithms utilising manually annotated datasets is likely to be required.

    • Stage 2:Machine learning as a model

    To test current developmental theories using annotated data from Stage 1, the student will explore hypothesis-driven questions such as:

  • Is there a difference in egocentric views across groups (parents/children without motor difficulties/children with motor difficulties)?
  • If yes to (1.),
  • Which group generates egocentric views that are most optimal for learning?
  • What properties of egocentric views are contributing to this difference?
  • Can egocentric view properties be predicted from motor constraints?

    • Supervisory team

    Dr Hana D’Souza - https://profiles.cardiff.ac.uk/staff/dsouzah

    Dr Marco Palombo - https://profiles.cardiff.ac.uk/staff/palombom

    Prof Yukun Lai - https://profiles.cardiff.ac.uk/staff/laiy4

    Research environment and training and development opportunities

    The student will benefit from interdisciplinary training, being jointly based in the Cardiff University Centre for Human Developmental Science (CUCHDS) and the Visual Computing Group. The lively broader research environment at Cardiff University (e.g., seminars, journal clubs, and workshops) will enable the student to form a network of future collaborators across disciplines.

    Responsible research and innovation

    Understanding typical/atypical human development and reducing tool bias is socially desirable and in the public interest. The dataset used in this project and the project goals are built on principles of equality, diversity, and inclusion.

    For any queries about this studentship, please contact the supervisor by email - Dr Hana D’Souza ([email protected] )

    Funding Comment

    This studentship is open to Home, EU or international students. The award offered will cover full fees and maintenance stipend.

    Funding Comment

    The studentship commences in October 2024, covers 3.5 years tuition fees and maintenance, with submission deadline of 4 years. The 2023/4 full-time maintenance grant was £18,662 p.a. Psychology students receive conference and participant money (~£2,250), computer, office space, access to courses and become members of the Doctoral Academy.


    Similar Positions