Cerebellar learning using perturbations

Abstract

The cerebellum aids the learning of fast, coordinated movements. According to current consensus, erroneously active parallel fibre synapses are depressed by complex spikes signalling movement errors. However, this theory cannot solve the credit assignment problem of processing a global movement evaluation into multiple cell-specific error signals. We identify a possible implementation of an algorithm solving this problem, whereby spontaneous complex spikes perturb ongoing movements, create eligibility traces and signal error changes guiding plasticity. Error changes are extracted by adaptively cancelling the average error. This framework, stochastic gradient descent with estimated global errors (SGDEGE), predicts synaptic plasticity rules that apparently contradict the current consensus but were supported by plasticity experiments in slices from mice under conditions designed to be physiological, highlighting the sensitivity of plasticity studies to experimental conditions. We analyse the algorithm’s convergence and capacity. Finally, we suggest SGDEGE may also operate in the basal ganglia.

Complex spike clusters and false‐positive rejection in a cerebellar supervised learning rule

“…An earlier study (Bouvier et al . ) reported LTD induction by two CS separated by 100 ms. In those experiments, the first CS coincided with granule cell activity and was interpreted as a perturbation signal, whereas the second, delayed CS acts as an error signal, with an overall PF–CF stimulus duration lasting 100 ms, a pattern resembling classic PF–CF coactivation protocols.…”

Motor context dominates output from Purkinje cell functional regions during reflexive visuomotor behaviours

“…In a developmental context, the cerebellum may be actively engaged in refining and maintaining sensorimotor behaviors as the physiology of neural circuits, muscles, and sensory appendages matures. In the context of supervised cerebellar learning, classical theories predict that the coincident activation of a climbing fiber input and parallel fiber synapses drives long-term 1120 depression at the active parallel fiber to Purkinje cell synapses, leading to motor learning (Ito, 2001; but see Bouvier et al, 2018). Synaptic plasticity mechanisms both at other synapses and involving other cerebellar neurons (e.g.…”

Depressed by Learning—Heterogeneity of the Plasticity Rules at Parallel Fiber Synapses onto Purkinje Cells

“…However, there are some notable exceptions (see Table 1, seventh column). Whereas some studies [83,88,89] suggest that LTD can only be induced in the absence of GABAergic inhibition, successful LTD induction has been reported without a block of GABAergic inhibition (Table 1, [4,60,82,112]). Some investigators have suggested that the study of LTD with inhibition blocked is “unphysiological”, however, another interpretation is that in vivo , the induction of LTD requires a transient reduction in the level of inhibitory input to the Purkinje cell from the molecular layer interneurons.…”

“…The synaptic physiologist must make many choices, as outlined above. There have been attempts to make choices that recreate some of the specific conditions that are present in vivo [52,68,112]. However, our lack of knowledge about what exactly these conditions are during learning remains a challenge.…”