463
个编辑
更改
跳到导航
跳到搜索
第1,068行：
第1,068行： − + 第1,074行：
第1,074行： − + 第1,082行：
第1,082行： − 除了整体的准确性，调查人员经常报告灵敏度和特异度，分别意味着真正阳性率(TPR)和真正阴性率(TNR)。同样，研究人员有时报告假阳性率(FPR)以及假阴性率(FNR)。然而，这些比率没有显示出它们的分子和分母。总操作特征是表达模型诊断能力的有效方法。目标为本课程显示上述比率的分子及分母，因此，目标为本课程提供的资料较常用的 ROC曲线(ROC)及 ROC 相关的曲线下面积(AUC)为多。+ − + 第1,092行：
第1,092行： − 机器学习引发了一系列伦理问题。经过数据收集和偏见训练的系统在使用时可能会表现出这些偏见(算法偏见) ，从而将文化偏见数字化。例如，使用具有种族主义招聘政策的公司的招聘数据，可能会导致机器学习系统重复这种偏见，对应聘者进行打分，以对照先前成功应聘者的相似程度。因此，负责任地收集系统使用的算法规则的数据和文档是机器学习的关键部分。+ 第1,100行：
第1,100行： − 因为人类语言包含偏见，训练语言语料库的机器也必然会学习这些偏见。+ 第1,108行：
第1,108行： − 其他形式的伦理挑战，与个人偏见无关，更多地出现在医疗保健领域。卫生保健专业人员担心，这些系统的设计可能不是为了公众的利益，而是作为创收机器。在美国尤其如此，美国长期存在一个道德两难境地，既要改善医疗保健，又要增加利润。例如，算法可以被设计为给病人提供不必要的测试或药物，而算法的所有者持有这些药物的股份。在医疗保健领域，机器学习有着巨大的潜力，可以为专业人员提供诊断、治疗甚至为病人规划康复路径的良好工具，但是在前面提到的个人偏见和这些“贪婪”偏见得到解决之前，这一切都不会发生。+ − + 第1,122行：
第1,122行： − + 第1,180行：
第1,180行： − + 第1,192行：
第1,192行： − + 第1,242行：
第1,242行： − + 第1,251行：
第1,251行： + +
→Journals
特别是机器学习方法可能会受到不同数据偏差的影响。仅针对当前客户的机器学习系统可能无法预测培训数据中没有表示的新客户群体的需求。当机器学习接受人造数据的训练时，很可能会挑选出社会中已经存在的同样的宪法和无意识的偏见。从数据中学到的语言模型已经被证明包含了类似人类的偏见。用于犯罪风险评估的机器学习系统被发现对黑人有偏见。在2015年，谷歌照片经常把黑人标记为大猩猩，而在2018年，这个问题仍然没有得到很好的解决，但据报道，谷歌仍然在使用变通方法从训练数据中删除所有大猩猩，因此根本无法识别真正的大猩猩。在许多其他系统中也发现了识别非白人的类似问题。2016年，微软测试了一个从 Twitter 上学来的聊天机器人，它很快就学会了种族主义和性别歧视的语言。由于这些挑战，机器学习的有效应用可能需要更长的时间才能在其他领域内广泛应用。人工智能科学家越来越关注机器学习中的公平问题，即减少机器学习中的偏见，推动机器学习为人类的利益服务。李等人提醒工程师们：“人工智能没有任何人为的东西... ... 它受到人的启发，由人创造，最重要的是，它会影响人。”，”这是一个强大的工具，我们才刚刚开始理解，这是一项意义深远的责任。”
特别是机器学习方法可能会受到不同数据偏差的影响。仅针对当前客户的机器学习系统可能无法预测培训数据中没有表示的新客户群体的需求。当机器学习接受人造数据的训练时，很可能会挑选出社会中已经存在的同样的宪法和无意识的偏见。从数据中学到的语言模型已经被证明包含了类似人类的偏见。用于犯罪风险评估的机器学习系统被发现对黑人有偏见。在2015年，谷歌照片经常把黑人标记为大猩猩，而在2018年，这个问题仍然没有得到很好的解决，但据报道，谷歌仍然在使用变通方法从训练数据中删除所有大猩猩，因此根本无法识别真正的大猩猩。在许多其他系统中也发现了识别非白人的类似问题。2016年，微软测试了一个从 Twitter 上学来的聊天机器人，它很快就学会了种族主义和性别歧视的语言。由于这些挑战，机器学习的有效应用可能需要更长的时间才能在其他领域内广泛应用。人工智能科学家越来越关注机器学习中的公平问题，即减少机器学习中的偏见，推动机器学习为人类的利益服务。李等人提醒工程师们：“人工智能没有任何人为的东西... ... 它受到人的启发，由人创造，最重要的是，它会影响人。”，”这是一个强大的工具，我们才刚刚开始理解，这是一项意义深远的责任。”
== Model assessments ==
== 模型评估 Model assessments ==
Classification machine learning models can be validated by accuracy estimation techniques like the [[Test set|Holdout]] method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-[[Cross-validation (statistics)|cross-validation]] method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, [[Bootstrapping|bootstrap]], which samples n instances with replacement from the dataset, can be used to assess model accuracy.<ref>{{cite journal|last1=Kohavi|first1=Ron|title=A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection|journal=International Joint Conference on Artificial Intelligence|date=1995|url=http://web.cs.iastate.edu/~jtian/cs573/Papers/Kohavi-IJCAI-95.pdf}}</ref>
Classification machine learning models can be validated by accuracy estimation techniques like the [[Test set|Holdout]] method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-[[Cross-validation (statistics)|cross-validation]] method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, [[Bootstrapping|bootstrap]], which samples n instances with replacement from the dataset, can be used to assess model accuracy.<ref>{{cite journal|last1=Kohavi|first1=Ron|title=A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection|journal=International Joint Conference on Artificial Intelligence|date=1995|url=http://web.cs.iastate.edu/~jtian/cs573/Papers/Kohavi-IJCAI-95.pdf}}</ref>
Classification machine learning models can be validated by accuracy estimation techniques like the Holdout method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-cross-validation method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, bootstrap, which samples n instances with replacement from the dataset, can be used to assess model accuracy.
Classification machine learning models can be validated by accuracy estimation techniques like the Holdout method, which splits the data in a training and test set (conventionally 2/3 training set and 1/3 test set designation) and evaluates the performance of the training model on the test set. In comparison, the K-fold-cross-validation method randomly partitions the data into K subsets and then K experiments are performed each respectively considering 1 subset for evaluation and the remaining K-1 subsets for training the model. In addition to the holdout and cross-validation methods, bootstrap, which samples n instances with replacement from the dataset, can be used to assess model accuracy.
分类机器学习模型可以通过精度估计技术进行验证，如 Holdout 方法，该方法将数据分成训练集和测试集(传统的2 / 3训练集和1 / 3测试集指定) ，并评估训练模型在测试集上的性能。对比分析表明，k 折叠交叉验证方法将数据随机划分为 k 子集，然后分别考虑评价子集和训练模型的剩余 K-1子集进行 k 实验。除了拒绝法和交叉验证方法，bootstrap 可以用来评估模型的准确性，它从数据集中取样 n 个实例并进行替换。
分类机器学习模型可以通过精度估计技术进行验证，如 Holdout 方法，该方法将数据分成训练集和测试集(传统的2 / 3训练集和1 / 3测试集指定) ，并评估训练模型在测试集上的性能。对比分析表明，'''k 折叠交叉验证方法 K-fold-cross-validation method'''将数据随机划分为 k 子集，然后分别考虑评价子集和训练模型的剩余 K-1子集进行 k 实验。除了拒绝法和交叉验证方法，bootstrap 也可以用来评估模型的准确性，它从数据集中取样 n 个实例并依次替换进行验证。
In addition to overall accuracy, investigators frequently report sensitivity and specificity meaning True Positive Rate (TPR) and True Negative Rate (TNR) respectively. Similarly, investigators sometimes report the False Positive Rate (FPR) as well as the False Negative Rate (FNR). However, these rates are ratios that fail to reveal their numerators and denominators. The Total Operating Characteristic (TOC) is an effective method to express a model's diagnostic ability. TOC shows the numerators and denominators of the previously mentioned rates, thus TOC provides more information than the commonly used Receiver Operating Characteristic (ROC) and ROC's associated Area Under the Curve (AUC).
In addition to overall accuracy, investigators frequently report sensitivity and specificity meaning True Positive Rate (TPR) and True Negative Rate (TNR) respectively. Similarly, investigators sometimes report the False Positive Rate (FPR) as well as the False Negative Rate (FNR). However, these rates are ratios that fail to reveal their numerators and denominators. The Total Operating Characteristic (TOC) is an effective method to express a model's diagnostic ability. TOC shows the numerators and denominators of the previously mentioned rates, thus TOC provides more information than the commonly used Receiver Operating Characteristic (ROC) and ROC's associated Area Under the Curve (AUC).
除了整体的准确性，调查人员经常报告灵敏度和特异度，分别意味着真正率 TPR 和真负率 TNR。同样，研究人员有时报告假正率 FPR以及假负率 FNR。然而，这些比率没有显示出它们的分子和分母。总操作特征是表达模型诊断能力的有效方法。目标为本课程显示上述比率的分子及分母，因此，较常用的有 ROC曲线及 ROC 相关的曲线下面积 AUC。
== Ethics ==
== 伦理问题 Ethics ==
Machine learning poses a host of [[Machine ethics|ethical questions]]. Systems which are trained on datasets collected with biases may exhibit these biases upon use ([[algorithmic bias]]), thus digitizing cultural prejudices.<ref>{{Cite web|url=http://www.nickbostrom.com/ethics/artificial-intelligence.pdf|title=The Ethics of Artificial Intelligence|last=Bostrom|first=Nick|date=2011|website=|access-date=11 April 2016|archive-url=https://web.archive.org/web/20160304015020/http://www.nickbostrom.com/ethics/artificial-intelligence.pdf|archive-date=4 March 2016|url-status=dead}}</ref> For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants.<ref name="Edionwe Outline">{{cite web|last1=Edionwe|first1=Tolulope|title=The fight against racist algorithms|url=https://theoutline.com/post/1571/the-fight-against-racist-algorithms|website=The Outline|accessdate=17 November 2017}}</ref><ref name="Jeffries Outline">{{cite web|last1=Jeffries|first1=Adrianne|title=Machine learning is racist because the internet is racist|url=https://theoutline.com/post/1439/machine-learning-is-racist-because-the-internet-is-racist|website=The Outline|accessdate=17 November 2017}}</ref> Responsible [[Data collection|collection of data]] and documentation of algorithmic rules used by a system thus is a critical part of machine learning.
Machine learning poses a host of [[Machine ethics|ethical questions]]. Systems which are trained on datasets collected with biases may exhibit these biases upon use ([[algorithmic bias]]), thus digitizing cultural prejudices.<ref>{{Cite web|url=http://www.nickbostrom.com/ethics/artificial-intelligence.pdf|title=The Ethics of Artificial Intelligence|last=Bostrom|first=Nick|date=2011|website=|access-date=11 April 2016|archive-url=https://web.archive.org/web/20160304015020/http://www.nickbostrom.com/ethics/artificial-intelligence.pdf|archive-date=4 March 2016|url-status=dead}}</ref> For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants.<ref name="Edionwe Outline">{{cite web|last1=Edionwe|first1=Tolulope|title=The fight against racist algorithms|url=https://theoutline.com/post/1571/the-fight-against-racist-algorithms|website=The Outline|accessdate=17 November 2017}}</ref><ref name="Jeffries Outline">{{cite web|last1=Jeffries|first1=Adrianne|title=Machine learning is racist because the internet is racist|url=https://theoutline.com/post/1439/machine-learning-is-racist-because-the-internet-is-racist|website=The Outline|accessdate=17 November 2017}}</ref> Responsible [[Data collection|collection of data]] and documentation of algorithmic rules used by a system thus is a critical part of machine learning.
Machine learning poses a host of ethical questions. Systems which are trained on datasets collected with biases may exhibit these biases upon use (algorithmic bias), thus digitizing cultural prejudices. For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants. Responsible collection of data and documentation of algorithmic rules used by a system thus is a critical part of machine learning.
Machine learning poses a host of ethical questions. Systems which are trained on datasets collected with biases may exhibit these biases upon use (algorithmic bias), thus digitizing cultural prejudices. For example, using job hiring data from a firm with racist hiring policies may lead to a machine learning system duplicating the bias by scoring job applicants against similarity to previous successful applicants. Responsible collection of data and documentation of algorithmic rules used by a system thus is a critical part of machine learning.
机器学习引发了一系列伦理问题。经过数据收集和偏见训练的系统在使用时可能会表现出这些偏见（算法偏差），从而将文化偏见数字化。例如，使用具有种族主义招聘政策的公司的招聘数据，可能会导致机器学习系统重复这种偏见，对应聘者进行打分，以对照先前成功应聘者的相似程度。因此，负责任地收集系统使用的算法规则的数据和文档是机器学习的关键部分。
Because human languages contain biases, machines trained on language corpora will necessarily also learn these biases.
Because human languages contain biases, machines trained on language corpora will necessarily also learn these biases.
因为人类语言包含偏见，训练语言语料库的机器也必然会学习到这些偏见。
Other forms of ethical challenges, not related to personal biases, are more seen in health care. There are concerns among health care professionals that these systems might not be designed in the public's interest but as income-generating machines. This is especially true in the United States where there is a long-standing ethical dilemma of improving health care, but also increasing profits. For example, the algorithms could be designed to provide patients with unnecessary tests or medication in which the algorithm's proprietary owners hold stakes. There is huge potential for machine learning in health care to provide professionals a great tool to diagnose, medicate, and even plan recovery paths for patients, but this will not happen until the personal biases mentioned previously, and these "greed" biases are addressed.
Other forms of ethical challenges, not related to personal biases, are more seen in health care. There are concerns among health care professionals that these systems might not be designed in the public's interest but as income-generating machines. This is especially true in the United States where there is a long-standing ethical dilemma of improving health care, but also increasing profits. For example, the algorithms could be designed to provide patients with unnecessary tests or medication in which the algorithm's proprietary owners hold stakes. There is huge potential for machine learning in health care to provide professionals a great tool to diagnose, medicate, and even plan recovery paths for patients, but this will not happen until the personal biases mentioned previously, and these "greed" biases are addressed.
而其他形式的伦理问题，与个人偏见无关，更多地出现在医疗保健领域。卫生保健专业人员担心，这些系统的设计可能不是为了公众的利益，而是作为创收机器。在美国尤其如此，美国长期存在一个道德两难境地，既要改善医疗保健，又要增加利润。例如，算法可以被设计为给病人提供不必要的测试或药物，而算法的所有者持有这些药物的股份。在医疗保健领域，机器学习有着巨大的潜力，可以为专业人员提供诊断、治疗甚至为病人规划康复路径的良好工具，但是在前面提到的个人偏见和这些“贪婪”偏见得到解决之前，这一切都不会发生。
== Software ==
== 相关软件 Software ==
[[Software suite]]s containing a variety of machine learning algorithms include the following:
[[Software suite]]s containing a variety of machine learning algorithms include the following:
=== Free and open-source software{{anchor|Open-source_software}} ===
=== 开源免费软件 Free and open-source software{{anchor|Open-source_software}} ===
{{Div col|colwidth=18em}}
{{Div col|colwidth=18em}}
=== Proprietary software with free and open-source editions ===
=== 具有免费和开源版本的专有软件 Proprietary software with free and open-source editions ===
{{Div col|colwidth=18em}}
{{Div col|colwidth=18em}}
=== Proprietary software ===
=== 专有软件 Proprietary software ===
{{Div col|colwidth=18em}}
{{Div col|colwidth=18em}}
== 杂志 Journals ==
== 相关期刊 Journals ==
* ''[[机器学习研究杂志 Journal of Machine Learning Research]]''
* ''[[机器学习研究杂志 Journal of Machine Learning Research]]''
* [[神经计算（期刊） Neural Computation (journal)|''Neural Computation'']]
* [[神经计算（期刊） Neural Computation (journal)|''Neural Computation'']]
== 相关会议 Conferences ==
== 相关会议 Conferences ==