Paper: Latent Predictive Learning

Our paper “The combination of Hebbian and predictive plasticity learns invariant object representations in deep sensory networks” is now published in Nature Neuroscience.

Sensory networks in our brain represent environmental objects as points on neural manifolds. The structure of these manifolds determines the abstractions communicated to other brain areas and, ultimately, how we perceive the world. In our article describe how synaptic plasticity can leverage the temporal structure in our sensory experience to shape representational manifolds and thereby create meaningful abstractions. This image uses cats and dogs on a neural manifold as an example, symbolizing the intricate dance between changes in cortical connectivity and neural representations that form abstractions in the brain. The metronome in the foreground represents time, which orchestrates this dance through the central notion of learning through prediction.

Artwork: Koshika Yadava, Skala Art. Background image: Cortical slice. Courtesy by Beatriz Lucia Saenz de Buruaga Corrochano, Friedrich Miescher Institute for Biomedical Research.