机器学习系统设计系统评估标准

VIP内容

论文题目

Model Cards for Model Reporting

论文摘要

训练有素的机器学习模式越来越多地用于执行执法、医学、教育和就业等领域的高影响力任务。为了澄清机器学习模型的预期用例,并尽量减少它们在不太适合的上下文中的使用,我们建议发布的模型附带详细说明其性能特征的文档。在本文中,我们提出了一个框架,我们称之为模型卡,以鼓励这种透明的模型报告。模型卡是经过培训的机器学习模型附带的简短文档,在各种条件下提供基准评估,例如跨不同文化、人口或表型群体(例如种族、地理位置、性别、Fitzpatrick皮肤类型)和跨部门群体(例如年龄和种族,或者性别和菲茨帕特里克皮肤类型)与预期应用领域相关。模型卡还披露了模型的使用环境、性能评估程序的细节以及其他相关信息。虽然我们主要关注以人为中心的机器学习模型在计算机视觉和自然语言处理领域的应用,但是这个框架可以用来记录任何经过训练的机器学习模型。为了巩固这一概念,我们为两种监督模式提供卡片:一种是训练来检测图像中的笑脸,另一种是训练来检测文本中的有毒评论。我们建议将模型卡作为机器学习和相关人工智能技术负责任民主化的一个步骤,提高人工智能技术如何工作的透明度。我们希望这项工作能够鼓励那些发布经过培训的机器学习模型的人在发布模型时附带类似的详细评估数字和其他相关文档。

论文作者

玛格丽特·米切尔、西蒙妮·吴、安德鲁·扎尔迪瓦尔、帕克·巴恩斯、露西·瓦瑟曼、本·哈钦森、埃琳娜·斯皮策、伊诺鲁瓦·德博拉·拉吉、蒂姆尼·格布鲁,来自google人工智能团队。

成为VIP会员查看完整内容
0
10

最新论文

COVID-19 which has spread in Iran from February 19, 2020, has infected 202,584 people and killed 9,507 people until June 20, 2020. The immediate suggested solution to prevent the spread of this virus was to avoid traveling around. In this study, the correlation between traveling between cities with new confirmed cases of COVID-19 in Iran is demonstrated. The data, used in the study, consisted of the daily inter-state traffic, air traffic data, and daily new COVID-19 confirmed cases. The data is used to train a regression model and voting was used to show the highest correlation between travels made between cities and new cases of COVID-19. Although the available data was very coarse and there was no detail of inner-city commute, an accuracy of 81% was achieved showing a positive correlation between the number of inter-state travels and the new cases of COVID-19. Consequently, the result suggests that one of the best ways to avoid the spread of the virus is limiting or eliminating traveling around.

0
0
下载
预览
父主题
Top