CRCNS: Common algorithmic strategies used by the brain for labeling points in high-dimensional space

这项工作的第一个主要目标是了解某些大脑区域（嗅觉系统， 海马体和小脑）学习非常复杂刺激，这些刺激采用组合编码， 将刺激识别为高维空间中的点。例如，简单的果蝇 嗅觉系统使用50种不同类型的气味受体的发射率， 通过将每种气味放置在SO维空间中的一个点来识别它。虽然苍蝇的嗅觉 系统已经被很好地理解，对脊椎动物大脑中的类似区域知之甚少，我们的研究 我们的目标是开始了解其他地区。第一步从鼠标开始 嗅觉系统在某些方面类似于苍蝇，但具有复杂性， 昆虫中没有。这些复杂性包括增强的处理噪声的能力， 气味并随着时间学习区分非常相似的气味（例如，两种类型的 红葡萄酒）。初步证据表明，应该可以了解 脊椎动物嗅觉的复杂性。研究设计包括研究 解剖学和记录放电率的不同类型的神经元在不同层次的 小鼠嗅觉系统和应用计算方法和算法 在早期的工作中被证明是成功的，以描述这些复杂性。 第二个主要目标是利用对这些大脑区域如何运作的洞察来改善 计算机算法的功能。长期以来，许多神经科学家和计算机科学家的梦想 科学家们一直在努力了解大脑是如何工作的， 来改善机器的运算能力事实上，经验表明，大脑 已经进化出计算机使用的信息处理算法的新变体 科学家解决一般的计算问题。对算法有足够的了解 这些洞察力可能会提供意想不到的方法来改善计算机的功能， 科学算法此外，了解所涉及的电路机制 在嗅觉处理中的作用可以帮助阐明各种嗅觉障碍的基础， 在未来会导致人工嗅觉装置的建造。

The first major goal of this work is to learn how certain brain regions (olfactory system, hippocampus, and cerebellum) learn very complex stimuli that employ a combinatorial code to identify stimuli as points in a high-dimensional space. For example, the simple fruit fly olfactory system uses the firing rates of 50 different types of odorant receptors to identify each odor by placing it at a point in a SO-dimensional space. Although the fly olfactory system is well understood, less is known about analogous regions in vertebrate brains, and our goal is to begin to learn about these other regions. The first step is to start with the mouse olfactory system that is similar to the fly in some ways but has complexities that are absent in insects. These complexities include an enhanced ability to handle noise in the odor and to learn over time to discriminate between very similar odors (e.g., two types of red wines). Preliminary evidence shows that it should be possible to learn the role of these complexities in vertebrate olfaction. The research design involves studying the anatomy and recording the firing rates of different types of neurons at different levels of the mouse olfactory system and in applying computational methods and algorithms that have proved successful in earlier work to describe these complexities. The second major goal is to use insights into how these brain regions operate to improve the function of computer algorithms. For a long time, a dream of many neuroscientists and computer scientists has been to understand how the brain works well enough that we could translate insights from the brain to improve machine computation. Indeed, experience has shown that the brain has evolved novel variations of information processing algorithms used by computer scientists to solve general computational problems. With sufficient insight into algorithms used by the brain, these insights may provide unexpected ways to improve the function computer science algorithms. Further, understanding the circuit mechanisms involved in olfactory processing can help illuminate the basis of a variety of smell disorders, and may in the future lead to the construction of artificial smelling devices.