You hear the words “Machine Learning” and “Artificial Intelligence” tossed around all the time nowadays. News articles pop up about how Artificial Intelligence (AI) is being used for predictive policing. You hear how companies are rushing to implement Machine Learning (ML) into their products. But what are AI and ML? How does it work? Read on to find out.

如今，您经常听到“机器学习”和“人工智能”一词被抛弃。 出现有关如何将人工智能(AI)用于预测性警务的新闻文章。 您会听到公司如何急于在其产品中实施机器学习(ML)。 但是什么是AI和ML ？ 如何运作 ？ 请仔细阅读，找出答案。

有什么不同？ 这些是什么？ (What’s the Difference? What are They?)

In everyday life, we use ML interchangeably with AI, but they’re different! AI is the whole domain whereas ML is a specific field in AI. In other words, AI is to math as ML is to geometry.

在日常生活中，我们可以将ML与AI互换使用，但是它们是不同的！ AI是整个领域，而ML是AI中的特定领域 。 换句话说，AI就像数学一样，ML就像几何。

So what are they? AI is the act of getting a machine to do actions that typically require human intelligence. This is a broad definition because it’s the definition for the whole field. AI applications range from chatbots, to composing music, to designing airplane parts!

那是什么 人工智能是使机器进行通常需要人类智慧的 行为的行为 。 这是一个广义定义，因为它是整个领域的定义。 人工智能的应用范围从聊天机器人到音乐创作 ，再到飞机零件的设计！

How does AI do this? One approach is ML. Machine Learning processes give machines data, then have them learn automatically from that data to produce actions. This may seem simple at first, but actually programming a program to classify cats and dogs seems like an impossible task! But don’t worry , it’s possible, and you’ll learn how it works in the next section.

人工智能如何做到这一点？ 一种方法是ML。 机器学习过程会提供机器数据，然后让他们从该数据中自动学习以产生动作 。 乍一看这似乎很简单，但是实际上编写一个程序对猫和狗进行分类似乎是不可能的任务！ 但是不用担心，这是可能的，您将在下一部分中学习它的工作方式。

ML如何工作？ (How does ML work?)

So how are we going to get a program to process data, and then learn from it automatically? It seems like a daunting task, but there’s a simple process that all ML models go through in order to learn:

那么，我们如何获得一个程序来处理数据，然后自动从中学习呢？ 这似乎是一项艰巨的任务，但是所有ML模型都要经过一个简单的过程才能学习：

The Process:

流程：

1. Take your data (inputs) and plug them into a function (architecture).

   获取数据( 输入 )并将其插入功能( 体系结构 )。

2. The function will process the data, with the help of some internal values (parameters), to produce some predictions.

   该函数将在一些内部值( 参数 )的帮助下处理数据，以产生一些预测。

3. The predictions will compare to the actual answers (labels) and will be used to calculate a loss.

   这些预测将与实际答案( 标签 )进行比较，并用于计算损失。

4. This loss will go back and update the internal values of the function.

   该损失将返回并更新函数的内部值。

We run through this process a couple of times, constantly updating our parameters, and we end up with an ML model! This very process is what the YouTube algorithm, Tesla’s autos, and Alpha Go’s algorithm go through in order to recommend videos, drive autos and beat people at Go! Alright, but what exactly are parameters, architectures, losses and all this jargon? In the next section, we’ll define and discuss these terms.

我们运行了几次 ，不断更新参数， 最后得到了ML模型 ！ YouTube算法，特斯拉(Tesla)的汽车和Alpha Go(Alpha Go)的算法正是通过这个过程来推荐视频，驾驶汽车并在Go上击败人们！ 好的，但是参数，架构，损耗以及所有这些术语到底是什么？ 在下一节中，我们将定义和讨论这些术语。

ML的定义： (Definitions for ML:)

• Input: Data in the form of numbers. Ex: Images are broken up into their bits, where each has 3 channels (RGB) these numbers are inputted into the architecture.

  输入： 数字形式的数据 。 例：图像被分解成比特，每个比特具有3个通道(RGB)，这些数字被输入到体系结构中。

• Parameters: Numbers that are inside the architecture. These will interact with the inputs and each other through a bunch of mathematical operations (mainly just addition and multiplication but there could be other ones depending on the architecture). We update these numbers, and it is purely from these numbers that we can do so much with ML.

  参数：体系结构内部的数字 。 这些将通过一堆数学运算(主要只是加法和乘法，但根据架构可能还有其他运算) 与输入相互交互 。 我们更新这些数字，而仅凭这些数字，我们就可以使用ML做到很多。

• Architecture: Shell or framework of the ML model. Inputs and parameters populate it, along with other layers/features which depend on the type of architecture.

  体系结构：ML模型的外壳或框架 。 输入和参数以及取决于体系结构类型的其他层/功能填充其中。

• Predictions: Outputs of the model in the form of Numbers. Inputs are inputted into the architecture and then interact with the parameters and features to produce predictions.

  预测：以Number形式的模型输出 。 输入被输入到架构中，然后与参数和特征交互以产生预测。

• Labels: Correct answers to the inputs in the form of Numbers. These labels will compare with the predictions in order to calculate the…

  标签：以数字的形式正确回答输入。 这些标签将与预测进行比较，以计算...

• Loss: Function that calculates how far off the predictions were from the answers. There are many ways to do this and will vary from application to application.

  损失：该函数计算预测与答案之间的距离。 有许多方法可以做到这一点，并且因应用程序而异。

If you’re planning on delving into the field of ML, it’s important to understand these basic definitions, as they are the foundation to the whole field.

如果您打算深入研究ML领域，那么了解这些基本定义非常重要，因为它们是整个领域的基础。

Thanks for reading! (Special thanks to Eason Wu for proofreading and editing!)

谢谢阅读！ (特别感谢Eason Wu的校对和编辑！)

获得访问专家视图的权限- 订阅DDI Intel (Gain Access to Expert View — Subscribe to DDI Intel)