Neural Coding in Visual and Auditory Systems for Natural Stimuli: Mathematical Modeling based on Data

神经科学是对神经系统的研究，包括大脑、脊髓和周围神经系统。它的目标是定义和理解从分子到细胞再到行为的连续体。神经科学现在尤其重要，因为除了我们对神经系统的内在兴趣外，该领域一直在以前所未有的速度收集丰富的数据(例如，功能磁共振成像，自然刺激下的电极测量)，并可以受益于严谨的数学和数据分析工具。Pi Yu博士是加州大学伯克利分校的统计学教授，该大学在海伦·威尔斯神经科学研究所有一个非常强大的神经科学小组。她认识到数据驱动的数学建模在神经科学中的重要性，并在基于数据的数学建模方面非常有经验。IGMS赠款的东道主是海伦·威利斯神经科学研究所的杰克·加兰特教授和弗雷德里克·特尼森教授。他们分别研究猕猴和鸣禽的视觉和听觉神经系统。宿主将提供具有自然刺激的宝贵神经元记录数据，并为PI的IGMS研究提供数据收集过程。IGMS研究的目标是获取不可或缺的神经科学知识，并了解宿主实验室的可用数据，以构建具有生物学意义的数学模型，将自然刺激与动物和人类视觉和听觉通路不同层面的反应(神经元放电率)联系起来。众所周知，在生理和心理物理实验中，在较低水平的通路上的一些细胞是线性的，而在较高水平的通路上的其他细胞是非线性的。因此，线性和非线性刺激-反应模型都是相关的。基于主办实验室的数据，PI计划1)将gMDL等模型选择技术应用于线性神经编码模型，并与现有的线性方法进行比较：2)在Boosting和支持向量机的框架下从机器学习中构建具有生物意义的非线性模型；3)使用自然相位图像和复数小波变换，基于独立分量分析(ICA)选择具有生物学意义的特征，并将它们用于第(1)和(2)部分的线性或非线性模型中。作为IGMS授予的结果，PI将能够为统计/数学专业的学生提供一个独特的机会，攻读计算神经科学中的应用博士学位。加州大学伯克利分校拥有非常强大的神经科学和物理科学系，因此它在计算神经科学方面具有令人难以置信的潜力。这些系的教员希望开发一个正式的跨学科的理论/计算神经科学研究生培训计划。IGMS的拨款将使PI成为这个新计划的一部分，从而帮助培训神经科学和数学或计算神经科学的下一代研究人员。对计算神经科学的研究生的培训将使社会受益，既增进了我们对大脑的理解，也启发了先进的工程应用。特别是，研究结果将让我们深入了解自然刺激是如何在动物和人类的视觉和听觉系统中处理的。了解生物系统是如何处理自然刺激的，可能会启发新算法在工程应用中的发展，如图像/声音压缩、语音识别、助听器或视觉辅助设备和神经假肢的图像/声音预处理。此外，IGMS的拨款将为PI打开神经科学领域的大门，一方面为PI解决该领域的问题，另一方面为PI在机器学习和数据挖掘方面的研究注入新的思想。该IGMS项目由MPS多学科活动办公室(OMA)和数学科学部(DMS)联合支持。

Neuroscience is the study of the nervous system, including the brain, spinal cord, and peripheral nervous system. Its goal is to define and understand the continuum from molecule to cell to behavior. Neuroscience is especially important now because in addition to our intrinsic interest in the nervous system, the field has been collecting rich data at an unprecedented rate (e.g. fMRI, electrode measurements with natural stimuli), and can benefit from rigorous tools for mathematical and data analyses.The PI Dr. Yu is a Professor of Statistics at UC Berkeley which has a very strong neuroscience group atthe Helen Wills Neuroscience Institute. She recognizes the importance of data-driven mathematical modeling in neuroscience at this point in time, and is very experienced with mathematical modeling based on data.The hosts for the IGMS grant are Professors Jack Gallant and Frederic Theunissen at the Helen Willis Neuroscience Institute. They study the visual and auditory nervous systems of macaques and songbirds, respectively. The hosts will provide the invaluable neuron recording data with natural stimuli andprovide access to the data collection process for the IGMS research of the PI.The goal of the IGMS research is to acquire the indispensable neuroscience knowledge and understand available data from the host labs to construct biologically meaningfulmathematical models to relate natural stimuli and responses (neuron firing rates) at different layers of the visual and auditory pathways of animals and humans. It is well-known from physiological and psychophysical experiments that some cells at lower levels of the pathways are linear and other cells at higher levels of the pathways are non-linear. Hence both linear and nonlinear stimulus-response models are relevant. Based on data from the host labs, the PI plans to 1) apply model selection techniques such as gMDL in linear neural coding models, and compare with existing linear methods;2) construct biologically meaningful non-linearmodels in the frameworks of boosting and Support Vector Machines from machine learning;3) select biologically meaningful features based on Independent Component Analysis (ICA) using natural phase images and complex wavelet transforms, and use them in linear or non-linear models from parts(1) and (2).As a result of this IGMS grant, the PI will be able to provide a unique opportunity for students of statistics/mathematics to pursuit a Ph.D. with applications in computational neuroscience. UC Berkeley has an incredible potential for a program in computational neuroscience since it has very strong departments both in neuroscience and in the physical sciences. Members of the faculty from these departments want to develop a formal interdisciplinary graduate training program in theoretical/computational neuroscience.The IGMS grant will enable the PI to be part of this new program and therefore help train the next generation of researchers at the interface of neuroscience and mathematics, or computational neuroscience.The training of graduate students in computational neuroscience will benefit society both by advancing our understanding of the brain and by inspiring advanced engineering applications. In particular, the research results will give us insight on how natural stimuli are processed in the visual and auditory system of animals and humans. Understanding how biological systems process natural stimuli might inspire the development of novel algorithms in engineering applications for image/sound compression, speech recognition and image/sound pre-processing for hearing or vision aids and neural prosthetics.Moreover, the IGMS grant will open up the field of neuroscience for the PI to solve problems in the field on one hand and on the other hand infuse new ideas into the PI's research in machine learning and data mining.This IGMS project is jointly supported by the MPS Office of Multidisciplinary Activities (OMA) and the Division of Mathematical Sciences (DMS).