Topic: self-supervised representation learning

Spontaneous activity in young ferrets (p22) is organised into modular, low-dimensional patterns across multiple cortical areas [1]. In the following two weeks (p22-43), these activity patterns undergo a transition similar in all areas, and show a gradual increase in dimensionality in most areas, suggesting an expansion of representational capacity [2]. However, it is unclear what the functional role of having low-dimensional activity patterns at the initial stage of development is, since one would expect the limited dimensionality to restrict representational learning. In this work, we investi

Predicting sensory input responses to update an internal ``world model'' is a core principle in cognitive learning theories that try to explain mechanisms behind representation learning in the brain [1,2]. Recent machine learning methods have also adopted predictive world modeling for self-supervised representation learning [3,4,5,6]. In this work, we investigate the perceptual and visuospatial capabilities of a self-supervised predictive world model trained in a manner approximating the active nature of human visual perception. We sample a sequence of fixations from image saliency maps [7], e