Understanding cortical computation through the lens of joint-embedding predictive architectures
Mohammadi, A.G.*, Halvagal, M. S.*, & Zenke, F.
bioRxiv, 2025
Our theory viewing cortical representation learning as predictive SSL with a JEPA.
Publications
Implicit variance regularization in non-contrastive SSL
Halvagal, M. S.*, Laborieux, A.*, & Zenke, F.
Advances in Neural Information Processing Systems 36 (NeurIPS 2023), 2023
Our paper explaining why the BYOL and SimSiam methods don’t suffer from representation collapse.
The combination of Hebbian and predictive plasticity learns invariant object representations in deep sensory networks
Halvagal, M. S., & Zenke, F.
Nature Neuroscience, 2023
Our theory that Hebbian plasticity helps predictive SSL to prevent representational collapse.
Inferring subjective preferences on robot trajectories using EEG signals
Iwane, F., Halvagal, M. S., Iturrate, I., Batzianoulis, I., Chavarriaga, R., Billard, A., & Millán, J. D. R.
2019 9th International IEEE/EMBS Conference on Neural Engineering (NER), 2019
This paper presents a proof of concept that EEG error potentials from a BCI can be used to course-correct the motion of a robotic arm in real-time.
( paper )