Learning rules

We look for learning rules that are local enough for neural circuits to implement, yet powerful enough to build useful representations from ongoing experience. Our work links unsupervised visual experience to changes in object-recognition behavior and inferior temporal cortex, tests how well unsupervised algorithms reproduce human real-time and lifelong learning, and shows that visual familiarity reorganizes activity dynamics and stimulus coding across inhibitory and excitatory cells. We now ask which local signals can improve representations without labels, how plasticity remains stable during continuous and non-stationary experience, and which roles recurrence, feedback, and inhibition play in assigning credit across a hierarchy.