Inferring the function performed by a recurrent neural network, bioRxiv, 2019-04-05

AbstractA central goal in systems neuroscience is to understand the functions performed by neural circuits. Previous top-down models addressed this question by comparing the behaviour of an ideal model circuit, optimised to perform a given function, with neural recordings. However, this requires guessing in advance what function is being performed, which may not be possible for many neural systems. Here, we propose an alternative approach that uses recorded neural responses to directly infer the function performed by a neural network. We assume that the goal of the network can be expressed via a reward function, which describes how desirable each state of the network is for carrying out a given objective. This allows us to frame the problem of optimising each neuron’s responses by viewing neurons as agents in a reinforcement learning (RL) paradigm; likewise the problem of inferring the reward function from the observed dynamics can be treated using inverse RL. Our framework encompasses previous influential theories of neural coding, such as efficient coding and attractor network models, as special cases, given specific choices of reward function. Finally, we can use the reward function inferred from recorded neural responses to make testable predictions about how the network dynamics will adapt depending on contextual changes, such as cell death andor varying input statistics, so as to carry out the same underlying function with different constraints.

biorxiv neuroscience 0-100-users 2019

Training and inferring neural network function with multi-agent reinforcement learning, bioRxiv, 2019-04-05

AbstractA central goal in systems neuroscience is to understand the functions performed by neural circuits. Previous top-down models addressed this question by comparing the behaviour of an ideal model circuit, optimised to perform a given function, with neural recordings. However, this requires guessing in advance what function is being performed, which may not be possible for many neural systems. To address this, we propose a new framework for optimising a recurrent network using multi-agent reinforcement learning (RL). In this framework, a reward function quantifies how desirable each state of the network is for performing a given function. Each neuron is treated as an ‘agent’, which optimises its responses so as to drive the network towards rewarded states. Three applications follow from this. First, one can use multi-agent RL algorithms to optimise a recurrent neural network to perform diverse functions (e.g. efficient sensory coding or motor control). Second, one could use inverse RL to infer the function of a recorded neural network from data. Third, the theory predicts how neural networks should adapt their dynamics to maintain the same function when the external environment or network structure changes. This could lead to theoretical predictions about how neural network dynamics adapt to deal with cell death andor varying sensory stimulus statistics.

biorxiv neuroscience 0-100-users 2019

Large-scale death of retinal astrocytes during normal development mediated by microglia, bioRxiv, 2019-04-04

Naturally-occurring cell death is a fundamental developmental mechanism for regulating cell numbers and sculpting developing organs. This is particularly true in the central nervous system, where large numbers of neurons and oligodendrocytes are eliminated via apoptosis during normal development. Given the profound impact of death upon these two major cell populations, it is surprising that developmental death of another major cell type – the astrocyte – has rarely been studied. It is presently unclear whether astrocytes are subject to significant amounts of developmental death, or how it occurs. Here we address these questions using mouse retinal astrocytes as our model system. We show that the total number of retinal astrocytes declines by over 3-fold during a death period spanning postnatal days 5-14. Surprisingly, these astrocytes do not die by apoptosis, the canonical mechanism underlying the vast majority of developmental cell death. Instead, we find that microglia kill and engulf astrocytes to mediate their developmental removal. Genetic ablation of microglia inhibits astrocyte death, leading to a larger astrocyte population size at the end of the death period. However, astrocyte death is not completely blocked in the absence of microglia, apparently due to the ability of astrocytes to engulf each other. Nevertheless, mice lacking microglia showed significant anatomical changes to the retinal astrocyte network, with functional consequences for the astrocyte-associated vasculature leading to retinal hemorrhage. These results establish a novel modality for naturally-occurring cell death, and demonstrate its importance for formation and integrity of the retinal gliovascular network.

biorxiv neuroscience 0-100-users 2019

ezTrack An open-source video analysis pipeline for the investigation of animal behavior, bioRxiv, 2019-03-30

AbstractTracking small animal behavior by video is one of the most common tasks in the fields of neuroscience and psychology. Although commercial software exists for the execution of this task, commercial software often presents enormous cost to the researcher, and can also entail purchasing specific hardware setups that are not only expensive but lack adaptability. Moreover, the inaccessibility of the code underlying this software renders them inflexible. Alternatively, available open source options frequently require extensive model training and can be challenging for those inexperienced with programming. Here we present an open source and platform independent set of behavior analysis pipelines using interactive Python (iPythonJupyter Notebook) that researchers with no prior programming experience can use. Two modules are described. One module can be used for the positional analysis of an individual animal across a session (i.e., location tracking), amenable to a wide range of behavioral tasks including conditioned place preference, water maze, light-dark box, open field, and elevated plus maze, to name but a few. A second module is described for the analysis of conditioned freezing behavior. For both modules, a range of interactive plots and visualizations are available to confirm that chosen parameters produce results that conform to the user’s approval. In addition, batch processing tools for the fast analysis of multiple videos is provided, and frame-by-frame output makes aligning the data with neural recording data simple. Lastly, options for cropping video frames to mitigate the influence of fiberopticelectrophysiology cables, analyzing specified portions of time in a video, and defining regions of interest, can be implemented with ease.

biorxiv neuroscience 200-500-users 2019

 

Created with the audiences framework by Jedidiah Carlson

Powered by Hugo