Neural mechanisms and functional requirements of complex odour identification in the mammalian olfactory system

大脑必须解决的一项基本计算任务是高维感官数据的分析和分类。我们认为嗅觉系统，作为最古老的感觉系统，已经进化出一个规范的解决方案来解决这个问题。因此，本项目的目标是研究嗅觉模型系统中感觉数据的皮层处理。该项目不仅旨在使用生物学上合理的数据驱动网络模型分析和识别这种处理背后的神经元机制，而且还旨在使用机器学习算法量化嗅觉系统的分类性能。本项目将融合现有的两个神经网络模型（一个是嗅球模型，另一个是嗅皮层模型），以真实地反映嗅觉信息处理的两个最重要的结构，并探索嗅觉系统中的振荡活动。这是特别有趣的，因为振荡（特别是在θ和γ波段）被认为是皮层信息处理的关键。对嗅皮质模型的模拟表明，振荡活动的起源在于嗅球，而传统上，振荡皮层活动归因于嗅皮质本身内部复杂的相互作用。此外，先前的模拟还表明，嗅觉皮层的结构分离为不同的子网络，这也与传统观点相反，传统观点认为嗅觉系统是一个自联想记忆网络。因此，拟议的项目将侧重于对嗅觉系统的这两种结构-功能假设进行严格的、基于模型的测试。最后，该项目将解决嗅觉系统中更抽象的分类性能问题。该项目将探索嗅觉皮层中信息的表达和处理是否能改善对复杂气味的分类。为此，最先进的分类算法（即支持向量机）将被用于从模拟嗅球神经活动或模拟嗅皮层神经活动中对复杂气味进行分类。我们将对这两种分类进行比较，以了解嗅觉皮层的信息处理是否会导致分类能力的增强。同样，这将被用于嗅觉系统的结构-功能假设的测试。

A fundamental computational task the brain has to solve is the analysis and classification of high-dimensional sensory data. We suggest that the olfactory system, as the oldest sensory system, has evolved a canonical solution to this problem. Therefore, the goal of the project proposed here, is to investigate the cortical processing of sensory data in the model system olfaction. The project aims to not only analyse and identify the neuronal mechanisms underlying this processing, using a biologically plausible, data-driven network model, but, in addition, to quantify the classification performance of the olfactory system using machine learning algorithms.In the proposed project two existing neuronal network models (one of the olfactory bulb and one of the olfactory cortex) will be fused, in order to realistically represent the two most important structures of olfactory information processing and to explore the oscillatory activity in the olfactory system. This is of particular interest since oscillations (especially in the theta and gamma band) are thought to be crucial in cortical information processing. Simulations of the model of olfactory cortex have suggested that the origin of oscillatory activity lies within the olfactory bulb, whereas, traditionally, oscillatory cortical activity is ascribed to complex interactions within the olfactory cortex itself. Furthermore, previous simulations also suggest a structural segregation of the olfactory cortex into distinct subnetworks, which also stands in contrast to the traditional view, which sees the olfactory system as an autoassociative memory network. Therefore, the proposed project will focus on a rigorous, model-based test of these two structure-function hypotheses of the olfactory system.Finally, the project will address the more abstract question of classification performance in the olfactory system. The project will explore whether the representation and processing of information within the olfactory cortex improves the classification of complex odours. To this end, state-of-the-art classification alogrithms (i.e. support vector machines) will be used to classify complex odours from either simulated neural activity in the olfactory bulb or simulated neural activity from the olfactory cortex. These two classifications will be compared in order to see whether information processing in the olfactory cortex leads to enhanced classification. Again, this will be tested for both structure-function hypotheses of the olfactory system.