Research
Research here focuses on automated science: uncertainty-aware deep learning, colour and qualia drift, embodied motor control, and neural plasticity work that feeds autonomous discovery loops. These themes develop the models, evaluation protocols, and idea-generation rituals that our systems run.
The institute also publishes technical reports, theoretical frameworks, and field notes that encode the agentic stack’s evolving logic.
Research themes
Uncertainty in Deep Learning
Calibrated deep learning through adaptive regularisation, online resampling, and geometry-aware Bayesian posteriors.
Colour, Consciousness & Qualia Drift
Empiricist theories of qualia paired with six-fundamental colour experiments, agent diaries, and installations that let people feel new spectra.
Motor Control & Embodied RL
Hierarchical world models, cerebellar-inspired controllers, and perturbation studies that push adaptive behaviour in robots and virtual agents.
Neural Plasticity & Representation
Theory and experiments on how synapses and adaptation rules learn structure across cortex and motor circuits.
Publications
Featured
Cross-regularization: Adaptive Model Complexity through Validation Gradients
International Conference on Machine Learning (ICML)
2025
Featured
The Blue is Sky: Color Qualia as Learned Associative Structures
Association for the Scientific Study of Consciousness (ASSC)
2025
Featured
World Models as Reference Trajectories for Rapid Motor Adaptation
Conference on Neural Information Processing Systems (NeurIPS)
2025
Precise Bayesian Neural Networks
arXiv
Twin-Boot: Uncertainty-Aware Optimization via Online Two-Sample Bootstrapping
arXiv
An Empiricist Connectionist Theory of Consciousness and Qualia
Working paper
The Relational Brain: Toward a New Neuroaesthetics
Essay
World Models as Reference Trajectories for Rapid Motor Adaptation
International Conference on Learning Representations (ICLR) – Robot Learning Workshop
Learning what matters: Synaptic plasticity with invariance to second-order input correlations
PLOS Computational Biology
Correlation-invariant synaptic plasticity
arXiv
Nonlinear Hebbian learning as universal principle in receptive field development
PLOS Computational Biology