Evidence for a causal inverse model in an avian cortico-basal ganglia circuit

Abstract

Learning by imitation is fundamental to both communication and social behavior and requires the conversion of complex, nonlinear sensory codes for perception into similarly complex motor codes for generating action. To understand the neural substrates underlying this conversion, we study sensorimotor transformations in songbird cortical output neurons of a basal-ganglia pathway involved in song learning. Despite the complexity of sensory and motor codes, we find a simple, temporally specific, causal correspondence between them. Sensory neural responses to song playback mirror motor-related activity recorded during singing, with a temporal offset of roughly 40 ms, in agreement with short feedback loop delays estimated using electrical and auditory stimulation. Such matching of mirroring offsets and loop delays is consistent with a recent Hebbian theory of motor learning and suggests that cortico-basal ganglia pathways could support motor control via causal inverse models that can invert the rich correspondence between motor exploration and sensory feedback.lateral magnocellular nucleus of the anterior nidopallium | Hebbian learning | mirror neuron

Naturalistic stimulation drives opposing heterosynaptic plasticity at two inputs to songbird cortex

“…Strikingly, the direction of plasticity can be inverted by changing the relative timing between activation of the two pathways. Plasticity of the two inputs is symmetrical not around zero offset between stimulation of the two inputs, but instead is centered around a temporal offset (~−75 ms, HVC leading LMAN) that is near the lag observed in anaesthetized recordings 41 and near the in vivo estimate for vocal-motor mirroring in LMAN 42 .…”

“…Third, auditory-vocal mirroring experiments have raised the possibility that some HVC-X neurons fire identically timed activity during song production and auditory playback 47,48 , suggesting they may be temporally offset from the direct HVC-RA motor commands. In contrast, LMAN neurons have a longer vocal-motor mirroring latency 42 . These provide yet another possible source of temporal offsets between HVC-RA and LMAN-RA drive.…”

Two neural streams, one voice: Pathways for theme and variation in the songbird brain

“…A complete picture of the zebra finch vocal control network would include information about the auditory memory of a bird's own song and the auditory memory of the paternal song acquired during the initial 'sensory' phase of learning. Although a full discussion of this work is beyond the scope of this review, it is worth noting that HVC and LMAN also contain an auditory memory of a bird's own songrevealed in anesthetized or sleeping birds by neural responses to playback of a bird's own song (Margoliash and Konishi, 1985;Dave and Margoliash, 2000;Giret et al, 2014). However, during singing, HVC and LMAN switch into a 'motor only' mode, where neurons no longer show responsiveness to auditory stimuli (Leonardo, 2004;Kozhevnikov and Fee, 2007;Giret et al, 2014;Hamaguchi et al, 2014).…”

“…Although a full discussion of this work is beyond the scope of this review, it is worth noting that HVC and LMAN also contain an auditory memory of a bird's own songrevealed in anesthetized or sleeping birds by neural responses to playback of a bird's own song (Margoliash and Konishi, 1985;Dave and Margoliash, 2000;Giret et al, 2014). However, during singing, HVC and LMAN switch into a 'motor only' mode, where neurons no longer show responsiveness to auditory stimuli (Leonardo, 2004;Kozhevnikov and Fee, 2007;Giret et al, 2014;Hamaguchi et al, 2014). It is also worth noting that HVC activity is necessary for juveniles to encode an auditory memory of paternal song (Roberts et al, 2012), and A duration x entropy scatterplot shows 5 syllable clusters from a day of singing before (pre) and after (post) bilateral LMAN ablation in an adult male zebra finch.…”

Learning temporal integration from internal feedback

“…We have shown how such a scheme can work for the learning of an inverse model that produces an appropriate motor code for the eye position out of a saccade command. More generally, learning an inverse model based on internal or external feedback to set a target for the upcoming motor state is a strategy that may also underly imitation learning (Giret et al 2014). Internal feedback may also be provided by the recall of past errors from an error memory that helps to quicker adapt to motor perturbations (Herzfeld et al 2014).…”