脑科学的研究越来越需要统计科学。过去几年里，脑科学实验数据以空前的速度增加：高密度多电极可以深入到工作中的大脑内部，同时收集成百上千的神经元的放电活动。脑科学领域的主要技术难点已经不再是如何从更多的神经元里读取数据，而是如何更系统化地理解这些大型电生理数据，从而帮助脑科学家回答"大脑是怎么工作的"这一重要问题。描述神经元放电活动的电生理数据是随时间连续观测的离散数据。而每个神经元是否放电又受到多个因素的影响。这样的非平稳点过程数据，如果采用通常的分析，可能面临统计功效下降、过拟合等挑战。在前期研究和与实验学家长期合作的基础上，我们发现广义线性模型和状态空间模型为这一类数据提供了很好的建模基础和分析框架。本项目中，我们将从普通点过程向标值点过程拓展，建立相应的估计方法，给出对应的算法和程序代码，并且继续与实验学家紧密合作，加深统计方法与脑神经数据的交叉应用，共同实现关于大脑工作机理的新发现。

Neuroscience needs more statistics. The past few years have witnessed an explosion in experimental neural data, propelled by the advent of new recording technologies: for example, high-density multiple-electrode arrays can now delve deep into a working brain, and record the simultaneous spiking activity of thousands of neurons. To answer the ultimate question of "how the brain works", the major bottleneck in neuroscience no longer lies in simply collecting data from large neural populations, but also in making sense of this data. Neurophysiological data that describe the spiking activity of neurons are discrete-valued measurements observed continuously over time, and many factors contribute to the spiking of a neuron. Inferences based on standard methods developed for continuous-valued signals can lead to a reduction in the statistical power available in the data, model overfitting, etc. Previous work has found that statistical estimation and inference procedures for neural spiking data are most appropriately developed based on the theory of point processes. In this project, we will extend the simple point process models to marked point process models, in order to handle the challenges posed by high-density recording methods. We will also develop corresponding estimation methods and provide custom-written computer algorithms and code for implementation, and continue our existing collaboration with experimentalists. We believe that our project closes a critical gap between neuroscience and statistics, and builds an interdisciplinary bridge that will lead to new discoveries about "how the brain works".