Search bioRxiv⌕ Search

bioRxiv · 10.1101/2025.07.18.665446

Correctness is its own reward: bootstrapping error signals in self-guided reinforcement learning

Abstract

Reinforcement learning (RL) offers a compelling account of how agents learn complex behaviors by trial and error, yet RL is predicated on the existence of a reward function provided by the agents environment. By contrast, many skills are learned without external guidance, posing a challenge to RLs ability to account for self-directed learning. For instance, juvenile male zebra finches first memorize and then train themselves to reproduce the song of an adult male tutor through extensive practice. This process is believed to be guided by an internally computed assessment of performance quality, though the mechanism and development of this signal remain unknown. Here, we propose that, contrary to prevailing assumptions, tutor song memorization and performance assessment are subserved by the same neural circuit, one trained to predictively cancel tutor song. To test this hypothesis, we built models of a local forebrain circuit that uses contextual premotor signals to cancel tutor song auditory input via synaptic plasticity. After learning, excitatory projection neurons signaled mismatches between the tutor song and birds own performance, best matching experimental data when learning involved anti-Hebbian plasticity in recurrent interneurons. We also found that learning proceeds by both sharpening error sensitivity and minimizing responses to the tutor song. Finally, the error signals produced by this model can train a simple RL agent to replicate the spectrograms of adult bird songs. These results suggest that local learning via predictive cancellation suffices for bootstrapping error signals capable of guiding self-directed learning of natural behaviors. SignificanceMany species, including humans, learn new skills by trial and error. But in the absence of external rewards and punishments, individuals need a means of assessing the quality of their performance. Here, we model the process by which juvenile male zebra finches memorize the song of an adult tutor, showing how local changes in brain circuits responsible for hearing can give rise during this process to an internal error signal that later allows for self-guided learning.

Explore related subjects

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Gong, Z., Duarte, F., Mooney, R., Pearson, J.. 2025-07-23. Correctness is its own reward: bootstrapping error signals in self-guided reinforcement learning. https://doi.org/10.1101/2025.07.18.665446

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

Functional validation of allele-specific LMNB1 silencing in patient-derived astrocytes as a therapeutic option for Autosomal Dominant Leukodystrophy

Adult-onset Autosomal Dominant Leukodystrophy (ADLD) is a rare fatal leukodystrophy caused by increased LMNB1 gene dosage, most commonly resulting from duplication of the LMNB1 locus. Because ADLD is a gene dosage disorder, selective reduction of pathological LMNB1 expression represents a rational therapeutic strategy. Although allele-specific RNA interference has previously been shown to lower LMNB1 levels in patient-derived fibroblasts and directly reprogrammed neurons, its therapeutic effects have not been evaluated in disease-relevant human glial cells or using functional efficacy endpoints. Here, we established human induced pluripotent stem cell-derived astrocytes from ADLD patients as a human glial model in which to validate allele-specific LMNB1 silencing across molecular, cellular, and functional readouts. ADLD astrocytes recapitulated increased LMNB1 expression and characteristic nuclear abnormalities and displayed transcriptional alterations affecting extracellular matrix organization, calcium homeostasis, metabolism and RNA processing. Functionally, these cells also exhibited functional phenotypes suitable for therapeutic evaluation: astrocyte-conditioned medium impaired the viability of both murine and human oligodendroglial cultures, while conditioned-medium and direct astrocyte-seeding paradigms revealed impaired post-lesion myelin recovery in lysolecithin-treated cerebellar organotypic slices. Allele-specific LMNB1 silencing restored physiological LMNB1 levels, corrected nuclear abnormalities, attenuated astrocyte-mediated oligodendroglial toxicity, improved post-lesion myelin recovery, and was associated with selective transcriptional programs associated with extracellular support and cholesterol metabolism. Together, these findings provide molecular, cellular, and functional validation of allele-specific LMNB1 dosage correction in patient-derived human astrocytes and offer key support for LMNB1-lowering strategies in disease-relevant human glial cells.

neuroscience↗

Perceptual integration of multisensory haptic, visual, and auditory feedback for roughness discrimination in augmented reality

Understanding how our different senses interact to shape our perception is essential to design realistic and immersive virtual and augmented reality (VR/AR) experiences. The present study investigated how roughness perception can be modulated through haptic, visual, and auditory cues in AR using a vibrotactile wristband. Participants compared virtual textures varying in vibration frequency/amplitude, visual grain size, and friction sound. Results revealed strong linear relationships between stimulus parameters and perceived roughness, with haptic frequency and visual cues driving the highest discrimination performance. Adding non-informative sensory feedback reduced perceptual sensitivity, acting as noise. Individual differences emerged: participants who rated haptic as the easiest modality showed greater sensitivity to haptic variations, while visual-reliant participants performed better with visual cues. We conclude that roughness in AR can be systematically manipulated, but is vulnerable to perceptual interference from irrelevant inputs, where our work provides actionable insights for implementing optimized and adaptive AR/VR interfaces.

neuroscience↗

Structural and functional MRI signatures of Gambling Disorder: a case-control study

Gambling disorder (GD) is a behavioural addiction that may help identify addiction-related neural features without the direct neurobiological effects of a primary substance of dependence. We examined regional grey matter volume (GMV) and resting-state functional connectivity (rsFC) in the same well-characterised sample. Eighteen men with GD and 21 matched healthy controls underwent high-resolution structural and resting-state functional MRI. GMV was quantified across 214 cortical and subcortical regions, and seed-based rsFC analyses focused on striatal subdivisions and mesocorticolimbic regions. Group differences were evaluated using permutation testing and cluster-corrected mixed-effects modelling. GD was associated with lower GMV in the ventromedial prefrontal cortex, orbitofrontal regions and other cortical and subcortical areas, alongside higher GMV in a subset of limbic and default-mode regions. Participants with GD also showed lower connectivity between the limbic striatum and the hippocampus, thalamus and putamen. In exploratory analyses, somatomotor connectivity was positively associated with gambling severity (Problem Gambling Severity Index: Spearman's rho = 0.71, p = 0.003, false-discovery-rate-adjusted q = 0.016). Structural and functional findings overlapped spatially in regions associated with valuation, memory, reward and habit formation, but regional GMV did not mediate group differences in rsFC. These findings are broadly consistent with corticostriatal models of GD and identify candidate circuit-level differences for independent replication. Larger, more diverse and longitudinal samples are required to establish their reproducibility, temporal direction and clinical relevance.

neuroscience↗