The Short Course on the Application of Machine Learning for Automated Quantification of Behavior

机器学习在行为自动量化中的应用短期课程

• 批准号: 10420570
• 负责人: Ishmail John Abdus-Saboor
• 金额: $ 15.46万
• 依托单位: JACKSON LABORATORY
• 依托单位国家: 美国
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-04-01 至 2026-03-31
• 项目状态: 未结题
• 来源: https://reporter.nih.gov/project-details/10420570
• 关键词: Adoption; Algorithms; Animal Behavior; Animals; Automobile Driving; Awareness; Behavior; Behavioral; Behavioral Genetics; Biological; Biomedical Research; Collaborations; Complex; Computer Analysis; Computer Vision Systems; Computer software; Data; Data Analyses; Data Science; Data Set; Databases; Development; Disadvantaged; Disease; Education; Educational process of instructing; Educational workshop; Environment; Ethology; Experimental Designs; Explosion; Faculty; Fellowship; Fostering; Future Teacher; Generations; Genetic; Genetic Research; Genomics; Goals; Health; Human; Institution; Knowledge; Learning; Machine Learning; Measurement; Mentors; Methodology; Methods; Minority-Serving Institution; Modeling; Neurophysiology - biologic function; Neurosciences; Neurosciences Research; Participant; Performance; Persons; Positioning Attribute; Postdoctoral Fellow; Psychiatry; Public Health; Reproducibility; Research; Research Personnel; Resolution; Resources; Running; Schedule; Science; Scientist; Statistical Methods; Structure; Students; Supervision; Technology; The Jackson Laboratory; Time; Trainers Training; Training; Underrepresented Minority; Work; base; career; career development; computer science; data integration; data modeling; data streams; deep learning; design; experience; experimental study; gender minority; graduate student; improved; innovation; insight; learned behavior; learning materials; lectures; machine learning method; neural network; neuroethology; next generation; novel; programs; recruit; relating to nervous system; statistical learning; statistics; symposium; technology development; temporal measurement; tool; virtual

PROJECT SUMMARY/ABSTRACT Elucidating the mechanism and function of neural encodings and circuit dynamics has been a major challenge in neuroscience and behavioral analyses. However, quantitative behavior analysis has dramatically accelerated and improved with the implementation and application of new machine learning methods, including new deep learning-based methods to track animals at high temporal and spatial resolution. This technology has broad current and potential application that will impact a breadth of fields that have direct relevance and impact on studies of human health and disease, including the fields of neuroscience, behavior, genetics, psychiatry, and biomedicine. However, several roadblocks limit the widespread adoption of these tools and analyses. First, many tracking and behavior analysis packages require a high level of computational expertise and are thus limited in application to expert labs. Second, with high-resolution data streams, quantitating behavior requires new statistical tools and proper modeling of data. Since the application of machine learning to behavioral analyses is an emerging and key methodology, we recognize an unmet need for investigators in a variety of relevant fields to learn the fundamentals of its rigorous use. Thus, to train a new generation of interdisciplinary researchers at the interface of neuroscience, machine learning, and behavior, we propose to establish an annual 4-day workshop that brings together experts in quantitative behavior, computer vision, and experimental design to provide a practical introduction to the field of quantitative neuroethology and behavior: we propose the unique and timely interdisciplinary course The Short Course on the Application of Machine Learning for Automated Quantification of Behavior at the Jackson Laboratory (JAX). This Short Course will provide attendees (in-person and virtually) with; information on the state-of-the-art of machine learning based behavior quantitation, the fundamentals of behavior quantitation, hands-on workshops and data analysis, a forum for student-teacher interaction for networking, and training at the leading edge of computational ethology. Students will emerge from the course with the ability to: 1) design a high quality, adequately powered behavior experiment; 2) select and install a suitable platform for high-resolution analysis of animal behavior; 3) deploy a behavior data analysis strategy, including collecting new training datasets, training analysis software, and validating performance on held-out data; and 4) run workflows/pipelines that are necessary to analyze their data following extraction. To achieve this, we propose: Aim 1. To develop and deliver a 4-day workshop to train scientists on application of machine learning to animal behavior quantitation. Aim 2. To create an environment that will expand the field of quantitative behavior analysis by fostering idea generation, discussion, and collaboration to yield new discoveries, broader applications, and advance technology development. Aim 3. Foster the recruitment and development of diverse junior investigators in neuroscience, behavioral genetics, and quantitative analysis of animal behavior.

项目摘要/摘要 阐明神经编码和电路动力学的机制和功能一直是一个重大挑战 在神经科学和行为分析中。但是，定量行为分析已大大加速 并随着新机器学习方法的实施和应用，包括新的深层 基于学习的方法以高时空和空间分辨率跟踪动物。这项技术广泛 当前和潜在的应用将影响直接相关并影响的广度 人类健康和疾病的研究，包括神经科学，行为，遗传学，精神病学和 生物医学。但是，几个障碍限制了这些工具和分析的广泛采用。首先，很多 跟踪和行为分析包需要高水平的计算专业知识，因此受到限制 应用于专家实验室。第二，使用高分辨率数据流，定量行为需要新的 统计工具和数据的正确建模。由于机器学习在行为分析中的应用是 一种新兴和关键方法，我们认识到对各种相关领域中的调查人员的需求 学习严格使用的基本原理。因此，培训新一代的跨学科研究人员 神经科学，机器学习和行为的界面，我们建议建立一个每年4天 汇集了定量行为，计算机视觉和实验设计专家的研讨会 为了对定量神经人体体体学和行为的领域提供实际介绍：我们提出 独特而及时的跨学科课程有关机器学习应用的简短课程 杰克逊实验室（JAX）的行为自动量化。这门简短的课程将为与会者提供 （面对面和实际上）有关基于机器学习的行为定量的最新信息的信息， 行为量化，动手研讨会和数据分析的基础，学生教师论坛 网络的相互作用和计算伦理学的前沿培训。学生将从 具有以下能力的课程：1）设计高质量的，充分的动力行为实验； 2）选择和 安装合适的平台，用于对动物行为的高分辨率分析； 3）部署行为数据分析 策略，包括收集新培训数据集，培训分析软件以及验证绩效 保留数据； 4）运行提取后分析其数据所需的工作流程/管道。到 实现这一目标，我们建议：目标1。开发和开展为期4天的研讨会，以培训科学家的应用 机器学习到动物行为定量。目标2。创建一个将扩大领域的环境 定量行为分析通过促进思想产生，讨论和协作以产生新 发现，更广泛的应用和进步技术开发。目标3。培养招聘和 开发各种初级研究者在神经科学，行为遗传学和定量分析中的发展 动物行为。