Biology subjects

Fee, M. S.

Publications and source records attributed to Fee, M. S..

7 recordsLinked to original sources

bioRxiv2026

Millisecond-scale, single-neuron credit assignment in a songbird

Learning an adaptive behavior requires identifying which actions, in which contexts, lead to particular outcomes. This problem of credit assignment is fundamental to both biological and artificial learners1-3. Songbird vocal learning presents a particularly demanding credit assignment problem: singing is controlled by millisecond-precise activity of thousands of motor neurons4-7, but song quality is encoded by a diffuse, delayed dopamine signal with more than an order of magnitude less temporal precision8-12. It is unknown how precisely the songbird brain can drive changes in specific premotor neurons at precise times to improve song performance. Here we show that the cortico-basal ganglia circuit thought to underlie songbird vocal learning can learn with millisecond-scale temporal precision and single-neuron spatial precision. We find that playing disruptive auditory feedback contingent on the activity of a targeted neuron in a vocal variability-generating premotor nucleus at one time in the song causes adaptive changes in the activity of the targeted neuron with 3.2 ms temporal precision. Learned changes in firing rate are not observed in uncorrelated neighboring neurons, thus revealing single-neuron spatial precision. These findings challenge the prevailing view of dopamine-mediated reinforcement as slow and imprecise13-15, and redefine our understanding of the limits of credit assignment in the brain.

neuroscience↗

bioRxiv2026

Constraints for spatially and temporally precise learning in a neural circuit model of reinforcement learning

Reinforcement learning is a key means by which animals learn appropriate actions in a given context. A large body of work suggests that such learning depends on interactions between cortico-basal ganglia circuits and the midbrain dopaminergic system, yet the underlying circuit mechanisms and plasticity rules are not fully understood. Here we present a biologically plausible, multi-region neural circuit model of songbird vocal learning and map it onto the actor-critic framework of reinforcement learning. In this model, stochastic spiking activity in the cortico-basal ganglia pathway implements action selection and drives behavioral exploration, while the pathways driving midbrain dopaminergic signaling evaluate behavioral outcomes and support a reward prediction error based learning rule that approximates stochastic gradient ascent. The model achieves millisecond-scale precise learning that matches observed behavior. We further use the model to examine two fundamental constraints on biological reinforcement learning. First, dopaminergic reinforcement signals are temporally imprecise, which can cause interference between neurons controlling actions that occur close in time. Second, dopaminergic signals are spatially imprecise, which can cause interference between neurons controlling different aspects of behavior but receiving a common reinforcement signal. By jointly modeling the actor and critic components of the circuit, we show that fast updating of reward prediction is crucial for precise and efficient learning under both forms of interference, and the model predicts the experimentally observed timescale of reward prediction updating. These results suggest a circuit-level mechanism by which biological systems achieve reinforcement learning despite the temporal and spatial limitations of global neuromodulatory signals.

neuroscience↗

bioRxiv2025

The songbird basal ganglia connectome

The basal ganglia (BG) play an essential role in shaping vertebrate behavior, ranging from motor learning to emotions, but comprehensive maps of their canonical synaptic architecture are missing. In mammals, three main neuronal pathways through the BG have been described - the direct, indirect, and hyperdirect pathways - which together orchestrate many aspects of learning and behavior. In songbirds, cell types associated with striatal and pallidal components appear intermingled in a single basal ganglia nucleus, Area X, essential for song learning. This allows for the dense reconstruction of the entire circuit within a compact volume. Here, we introduce the first vertebrate basal ganglia connectome, comprising over 8,500 automated neuron reconstructions connected by about 20 million synapses. High image quality and automated reconstruction allowed analysis with minimal manual proofreading. Based on direct anatomical measurement of synaptic connectivity, we confirm that a direct, indirect and hyperdirect pathway can be traced through Area X. However, detailed morphological and connectomic analysis revealed no clearly distinct direct and indirect medium spiny neuron subpopulations, and a dominance of the direct and hyperdirect pathway. In addition to previously identified neuron types in Area X, we could distinguish three novel GABAergic neuron types, two of which are major output targets of GPe neurons, leading to novel feedback circuitry within Area X. We further found unexpectedly strong neuronal interconnectivity and recurrency between neurons associated with all pathways. Our data thus challenge the universality of the view of the basal ganglia as an information processor organized into discrete feedforward pathways.

neuroscience↗

bioRxiv2023

Spatial Tracking Across Time (ST AT ): Tracking Neurons Across In-Vivo Imaging Sessions through Optimizing Local Neighborhood Motion Consistency

Chronic calcium imaging has become a powerful and indispensable tool for analyzing the long-term stability and plasticity of neuronal activity. One crucial step of the data processing pipeline is to register individual neurons across imaging sessions, which usually extend over a few days or even months, and show various levels of spatial deformation of the imaged field of view (FOV). Previous solutions align FOVs of all sessions first and then register the same neurons according to their shapes and locations [1, 2]. However, the FOV registration is computational intensive, especially in the case of nonrigid case. Here we propose a cell tracking method that does not require FOV image registration. Specifically, the algorithm STAT (short for Stay T ogether, Align Together, and for Spatial Tracking Across Time) represents neurons from two sessions as two sets of neuronal centroids, uses point set registration (PSR) to find a spatially smooth transformation to align them while assigning correspondences. The optimization method iteratively updates between the general motion and individual neuron identity tracking, an idea seen in the computer vision literatures [3, 4]. Our method can be thought of as a specialization and simplification of these more general methods to calcium imaging neuron tracking. We validate STAT on datasets with simulated nonrigid motion that is hard to motion correct without extensive manual intervention. Next, we test STAT on experimental data from singing birds collected on three different days, and observe stable song-locked activity across days. An example use case of this package is reference [5].

neuroscience↗

bioRxiv2021

A Novel Optical Design Enabling Lightweight and Large Field-Of-View Head-Mounted Microscopes

We present a novel fluorescence microscope light path that enables imaging, during free behavior, of thousands of neurons in mice and hundreds of neurons in juvenile songbirds. The light path eliminates traditional illumination optics, allowing for head-mounted microscopes that have both a lower weight and a larger field-of-view (FOV) than previously possible. Using this light path, we designed two microscopes: one optimized for field-of-view ([~]4 mm FOV; 1.4 g), and the other optimized for weight (1.0 mm FOV; 1.0 g).

neuroscience↗

bioRxiv2020

An anatomical substrate of credit assignment in reinforcement learning

A key problem in learning is credit assignment. Biological systems lack a plausible mechanism to implement the backpropagation approach, a method that underlies much of the dramatic progress in artificial intelligence. Here, we use automated connectomic analysis to show that the synaptic architecture of songbird basal ganglia (Area X) supports local credit assignment using a variant of a node perturbation algorithm proposed in a model of reinforcement learning. Using two volume electron microscopy (vEM) datasets, we find that key predictions of the model hold true: axons that encode exploratory variability terminate predominantly on dendritic shafts, while axons that encode song timing (context) terminate predominantly on spines. Based on the detailed EM data, we then built a biophysical model of reinforcement learning that suggests that the synaptic dichotomy between variability and context encoding axons facilitates efficient learning. In combination, these findings provide strong evidence for a general, biologically plausible credit assignment model in vertebrate basal ganglia learning. One Sentence SummaryUsing automated connectomic analysis and biophysical modeling, we show how the basal ganglia could solve the credit assignment problem on the synaptic level.

neuroscience↗

bioRxiv2018

Unsupervised discovery of temporal sequences in high-dimensional datasets, with applications to neuroscience

Identifying low-dimensional features that describe large-scale neural recordings is a major challenge in neuroscience. Repeated temporal patterns (sequences) are thought to be a salient feature of neural dynamics, but are not succinctly captured by traditional dimensionality reduction techniques. Here we describe a software toolbox--called seqNMF--with new methods for extracting informative, non-redundant, sequences from high-dimensional neural data, testing the significance of these extracted patterns, and assessing the prevalence of sequential structure in data. We test these methods on simulated data under multiple noise conditions, and on several real neural and behavioral data sets. In hippocampal data, seqNMF identifies neural sequences that match those calculated manually by reference to behavioral events. In songbird data, seqNMF discovers neural sequences in untutored birds that lack stereotyped songs. Thus, by identifying temporal structure directly from neural data, seqNMF enables dissection of complex neural circuits without relying on temporal references from stimuli or behavioral outputs.

neuroscience↗