Machine Intelligence for Neuroscience Experimental Control

了解大脑及其产生的行为是我们这个时代的一项重大科学挑战。为了取得成功，科学家必须能够解释动物行为与不同大脑区域的神经活动之间的关系。这需要仔细设计和操作行为实验，实验人员记录或操纵神经活动，而动物（例如非人类灵长类动物，啮齿动物，鱼类，昆虫）从事需要仔细观察和量化的特定行为。行为和脑科学实验室的实验需要集成和控制来自多个记录设备（视频、神经活动测量电极、传感器）的硬件的软件，以及能够解释大型复杂行为和神经数据集的分析工具。研究大脑和行为的科学家将大部分时间用于设计实验和分析数据，而在数据获取本身上花费的时间最少，这可能会影响数据的质量。此外，全球数百个神经科学研究小组开发了自己的实验和分析工具，大多数使用不同的编程语言，导致数据共享和分析效率低下，并影响可重复性（即其他人重复相同实验的容易程度）。在这里，我们建议为科学界提供一个软件工具，将大大提高实验控制和数据分析的效率。我们将为现有的软件平台Bonsai开发一套新的功能。Bonsai是一个完全集成的软件环境，强调性能，灵活性和易用性，允许以前没有编程经验的科学家快速开发自己的高性能数据采集和实验控制系统。到目前为止，盆景已经被全球数百名科学家采用，为行为科学和脑科学提供交互式实验控制。在这个提案中，我们的目标是通过一个在线和离线机器智能工具工具箱来扩展盆景的功能，用于分析行为和神经数据集，并创建一个开放的软件共享平台。Bonsai的增强功能将使新型研究成为可能，并通过以下方式加速发现和提高效率：(1)为缺乏专业知识的实验室提供使用这些工具的途径；(2)减少在多个实验室中重新发明相同工具的需要；(3)标准化数据处理流，从而提高实验室之间的可重复性。我们相信这一努力将促成并加速大脑如何产生行为的新发现。

Understanding the brain and the behaviour it generates is a major scientific challenge of our era. To succeed, scientists must be able to explain how animal behaviour relates to neural activity across different brain regions. This requires careful design and manipulation of behavioural experiments, where experimenters either record or manipulate neural activity while the animals (e.g. non-human primates, rodents, fish, insects) engage in specific behaviours which need to be carefully observed and quantified. Experiments in behavioural and brain science laboratories require software that integrates and controls hardware from multiple recording devices (video, electrodes for neural activity measurement, sensors), and analysis tools that can interpret large and complex behavioural and neural datasets. Scientists studying brain and behaviour dedicate the majority of their time designing experiments and analysing the data, with least time spent on data acquisition itself, which may impact the quality of data. Moreover, hundreds of neuroscience research groups worldwide develop their own experimental and analytical tools, most using different programming languages, leading to inefficiencies in data sharing and analysis, and impacting reproducibility (i.e. how easy it is for someone else to repeat the same experiment). Here we propose to provide the scientific community with a software tool that will dramatically increase the efficiency of experimental control and data analysis. We will do so by developing a new set of functionalities to an existing software platform, Bonsai. Bonsai is a fully integrated software environment that emphasises performance, flexibility, and ease-of-use, allowing scientists with no previous programming experience to quickly develop their own high-performance data acquisition and experimental control systems.Thus far, Bonsai has been adopted by hundreds of scientists worldwide to provide interactive experimental control in behavioural and brain sciences. In this proposal, we aim to extend Bonsai's functionality with a toolbox of online and offline Machine Intelligence tools for analysis of behavioural and neural datasets, and to create an open-access platform for software sharing. Bonsai's enhanced functionality will enable new types of research, and speed up discovery and improve efficiency by (i) providing access to such tools to laboratories lacking expertise, (ii) reducing the need to reinvent the same tools in multiple labs and (iii) standardising the data processing streams, thus increasing reproducibility across laboratories. We believe this effort will enable and accelerate new discoveries in how the brain generates behaviour.