机器学习术语详解与TensorFlow专业词汇对照

需积分: 14 5 下载量 85 浏览量 更新于2024-07-17 收藏 1017KB PDF 举报
本文档是一份全面的机器学习术语表,涵盖了通用的机器学习概念以及TensorFlow平台特有的术语。以下是对文中提及的一些关键知识点的详细解释: 1. **A/B测试 (A/B Testing)**: 这是一种统计方法,用于比较两个或多个不同版本的技术效果,常用于评估新旧技术的性能。通过设定一个对照组(A组)和实验组(B组),在一定时间内收集数据,然后分析结果来确定哪个版本更有效。A/B测试不仅可以判断优劣,还能确认这种差异是否具有统计上的显著性。 2. **准确率 (Accuracy)**: 在分类任务中,准确率是模型正确预测样本的比例。在二元分类中,它是指预测为正类的样本数占总样本数的比例;而在多类别分类中,准确率则涉及到每一类别分别计算正确预测的比率,再取平均值。它不适用于不平衡的数据集,因为少数类别的正确预测可能占据较大权重。 3. **激活函数 (Activation Function)**: 是神经网络中的一种非线性转换函数,如ReLU(Rectified Linear Unit)和S型函数(sigmoid)。它们的作用是增加模型的表达能力,使神经元可以捕捉到输入之间的复杂关系,并为后续层提供输入。 4. **AdaGrad**: 一种自适应学习率优化算法,它根据历史梯度信息动态调整每个参数的学习率,使得在训练过程中对不同参数施加不同的学习速度,有助于提高模型的收敛性和适应性。 AdaGrad特别适合稀疏数据和在线学习。 5. **ROC曲线 (Receiver Operating Characteristic Curve) 和 AUC (Area Under the Curve)**: ROC曲线是衡量分类器性能的可视化工具,横轴表示假阳性率(False Positive Rate),纵轴表示真阳性率(True Positive Rate)。AUC表示整个ROC曲线下的面积,数值越大,表示分类器性能越好,AUC=1表示完美分类,而AUC=0.5表示随机猜测。 6. **反向传播算法 (Backpropagation)**: 用于训练神经网络的优化算法,通过前向传播计算输出值,然后从输出层逆向回溯,计算损失函数对每个权重的梯度,从而更新权重以最小化损失。这是深度学习训练的核心过程。 7. **基准 (Baseline)**: 常用于评估模型性能的起点或参照标准。在某些情况下,简单的策略或预设模型可能就是基线,新模型需要超越这个基准来证明其有效性。 这份术语表对于理解和应用机器学习特别是TensorFlow提供了基础参考,涵盖了从实验设计到模型训练的关键概念和技术细节。对于从事相关领域的研究者、开发者或学生来说,这是一个宝贵的参考资料。
2018-07-29 上传
Machine Learning with TensorFlow gives readers a solid foundation in machine-learning concepts plus hands-on experience coding TensorFlow with Python. Purchase of the print book includes a free eBook in PDF, Kindle, and ePub formats from Manning Publications. About the Technology TensorFlow, Google's library for large-scale machine learning, simplifies often-complex computations by representing them as graphs and efficiently mapping parts of the graphs to machines in a cluster or to the processors of a single machine. About the Book Machine Learning with TensorFlow gives readers a solid foundation in machine-learning concepts plus hands-on experience coding TensorFlow with Python. You'll learn the basics by working with classic prediction, classification, and clustering algorithms. Then, you'll move on to the money chapters: exploration of deep-learning concepts like autoencoders, recurrent neural networks, and reinforcement learning. Digest this book and you will be ready to use TensorFlow for machine-learning and deep-learning applications of your own. What's Inside Matching your tasks to the right machine-learning and deep-learning approachesVisualizing algorithms with TensorBoardUnderstanding and using neural networks About the Reader Written for developers experienced with Python and algebraic concepts like vectors and matrices. About the Author Author Nishant Shukla is a computer vision researcher focused on applying machine-learning techniques in robotics. Senior technical editor, Kenneth Fricklas, is a seasoned developer, author, and machine-learning practitioner. Table of Contents PART 1 - YOUR MACHINE-LEARNING RIGA machine-learning odysseyTensorFlow essentialsPART 2 - CORE LEARNING ALGORITHMSLinear regression and beyondA gentle introduction to classificationAutomatically clustering dataHidden Markov models PART 3 - THE NEURAL NETWORK PARADIGMA peek into autoencodersReinforcement learningConvolutional neural networksRecurrent neural networksSeque