August 11th, 2025, Amsterdam
3:00 - 6:00 pm - Room A2.11
3:00 - 6:00 pm - Room A2.11
From Child to Machine Learning: The potential and challenges of translating Developmental Principles into Neural Network Design
Background
The human visual system is full of optimisations—mechanisms designed to extract the most useful information from a constant stream of incoming data. The field of neuro-AI has made significant progress in aligning deep neural networks (DNNs) with adult human brains and behaviour, and has successfully captured some of these optimisations. However, less attention has been paid to deliberately designing networks whose learning trajectories mirror human cognitive development. Can addressing this gap offer new insights into how neural optimisations for visual learning unfold across evolutionary and developmental timelines?
Workshop
This workshop will begin with an overview of neural-network design and training in developmental psychology followed by experts talks on incorporating developmental principles into vision models, showcasing current methods and discussing key issues around curriculum learning approaches inspired by infant milestones, biologically motivated architectural constraints, and evaluation metrics that track alignment with human developmental benchmarks. In structured breakout sessions, participants will collaboratively develop research agendas and potential interdisciplinary projects, culminating in a panel discussion to synthesize insights and chart future directions for the field.
By bringing together experts from developmental psychology, cognitive neuroscience, and AI research, this workshop aims to advance methods for creating neural networks that develop capabilities in more human-like ways, offering fresh perspectives on why humans are such efficient and flexible learners.
Birkbeck, University of London
MIT
Trinity College Dublin
University of California, San Diego
Freie Universitat
University of Amsterdam
University College London
Key questions we aim to advance:
Learning Curricula; How can staged training protocols inspired by infant perceptual milestones, scaffold learning and boost model robustness? From unsupervised play to guided training, can learning regimes shape efficiency?
Architectural Constraints; How do wiring constraints of the developing brain shape how infants learn, and what does it mean to embed those biases into neural networks?
Evaluation Metrics; Moving beyond static snapshot performance metrics, toward concordance with human adaptability, resilience and developmental benchmarks
Preliminary Program:
15:00-15:05 – Introduction and outlining objectives
Dr. Tessa Dekker (Chair), University College London
15:05-15:25 – Keynote: "Developmental Principles as Blueprints for Neural Networks" Professor Denis Mareschal, Birkbeck, University of London
15:25-17:05 – Invited Panel of Expert Speakers (15 minutes + 5 mins Questions)
· Dr. Lukas Vogelsang, MIT
· Aine Dineen, Trinity College Dublin
· Dr Bria Long, University of California, San Diego
· Zejin Lu, Freie Universität
· Dr. Steven Scholte, University of Amsterdam
17:05-17:15 Integration & Overview of Key Challenges
17:15-17:45 Breakout groups
17:45-18:30 Panel Discussion, drinks & snacks