计算机知识78:深度学习,原来普通人也能搞懂深度学习,这个干货绝对能看明白!双语版本

深度学习，其实就是教电脑像我们人脑一样去看东西去想事儿对吧。 Deep learning is actually teaching computers to think and see things just like our human brains, right?

你平时刷手机刷到的人脸识别，还有那个能跟你聊天的AI，背后全是它在干活哦。 The face recognition you usually encounter when scrolling through your phone, and that AI that can chat with you, it's all working behind the scenes, okay?

它不是什么特别新的稀奇玩意儿，但就是最近这些年才突然火得一塌糊涂。 It's not some super new fancy stuff, but it has only become extremely popular in recent years.

它其实就是机器学习里面的一个分支，核心就是用多层的神经网络去摸清楚数据里面藏着的规律。 It's actually a branch of machine learning, and the core is using multi-layer neural networks to figure out the hidden patterns in the data.

就像我们小时候学认字，看多了你就记住哪个字长什么样了对吧？深度学习也是一样的，给它喂超级多的资料，它自己就慢慢学会分辨东西了。 It's just like when we learned to read characters as kids, after seeing enough of them you remember what each character looks like, right? Deep learning is the same, feed it a ton of data, and it will slowly learn to distinguish things by itself.

你说神不神奇啊？它不是程序员把规则一条一条写死给它的哦。是它自己从一堆乱七八糟的数据里面，自学成才搞出来的本事。 Is that not amazing? It's not like programmers write every single rule for it by hand. It learns all its skills by itself from a pile of messy data, self-taught.

之前好多AI方法，遇到稍微复杂一点的问题就直接歇菜了。 A lot of old AI methods would just fail completely when they encountered a slightly more complicated problem.

比如让电脑认出来一张照片里面是猫还是狗，之前的方法做出来准确率低得吓人，根本没法用。 For example, asking a computer to tell if a cat or a dog is in a photo, the accuracy of old methods was super low, they were totally unusable.

结果深度学习出来之后呢？准确率直接飙升，现在比人认的还准你敢信？ But after deep learning came out? The accuracy just shot straight up, can you believe it's even more accurate than humans now?

深度学习为什么能做到啊，你琢磨琢磨。 Why can deep learning do this, just think about it for a second.

它最牛的地方，就是能自动把数据里面的特征给提取出来，不用人再吭哧吭哧手动做了。 Its biggest strength is that it can automatically extract features from data, you don't need people to do it manually by sweating through all the work anymore.

之前做图像识别，得工程师自己想怎么提取边缘啊怎么找轮廓，费老鼻子劲了效果还不好。 Before doing image recognition, engineers had to figure out how to extract edges and find outlines by themselves, it took so much effort and still didn't work well.

现在好了，把数据往模型里面一丢，它自己就把该学的都学会了，省了多少事啊。 Now it's great, just throw the data into the model, it learns everything it needs to learn by itself, saves so much trouble.

还有啊，现在咱们能拿到的数据越来越多，电脑的计算能力也越来越强了，刚好撞上了深度学习的需求。 Also, now we have more and more data available, and the computing power of computers is getting stronger and stronger, which just fits exactly what deep learning needs。

深度学习就是越喂数据越厉害，你数据越多它就越聪明，刚好赶上现在大数据的风口，不火才奇怪呢。 Deep learning gets better the more data you feed it, the more data you have the smarter it gets, it just caught the big data boom, it would be weird if it wasn't popular.

而且你看现在多少应用全靠它啊，导航地图给你算最优路线，外卖APP给你推你爱吃的店，甚至你去医院拍CT，AI都能帮着找病灶呢，这些全是深度学习的功劳啊。 And look how many applications rely on it now, navigation maps calculate the best route for you, takeout apps recommend restaurants you love, even when you go to the hospital for a CT scan, AI can help find lesions, all of these are the work of deep learning.

好多人一听到深度学习这五个字，第一反应就是我数学不好我肯定学不会，直接就打退堂鼓了，至于吗？ When a lot of people hear the words deep learning, the first reaction is I'm bad at math I definitely can't learn it, they just give up right away, is that really necessary?

其实真没你想的那么吓人，哪怕你就是个普通爱好者，也能慢慢上手玩一玩啊。 It's really not as scary as you think, even if you're just a regular hobbyist, you can slowly get started and play around with it.

第一步当然得先搞懂最基础的概念啊，啥是神经网络，啥是神经元，这些最基本的得先搞明白对吧。 The first step is definitely to understand the most basic concepts, what is a neural network, what is a neuron, you have to figure these most basic things out first, right.

先从汉语的基础资料看起，别一上来就啃全英文的论文，那不是给自己找罪受吗。 Start with basic materials in Chinese first, don't jump straight into reading all-English papers, that's just asking for trouble.

先把神经网络的前向传播和反向传播这两个核心东西搞懂，不用一开始就去抠那些复杂的数学推导，先懂个大概意思就行。 First figure out the two core things forward propagation and backpropagation of neural networks, you don't need to dig into all the complicated math derivations at the start, just get a general idea first is enough.

然后就得动手敲代码啊，光看不练假把式对不对。 Then you have to start writing code, just reading without practicing is all fake skills, right.

现在网上好多现成的框架，啥TensorFlow啊PyTorch啊，直接拿来用就行，不用你自己从0开始写整个网络，多方便啊。 There are so many ready-made frameworks online, like TensorFlow and PyTorch, you can just use them directly, you don't need to write the whole network from zero yourself, how convenient is that.

先从最简单的手写数字识别开始做，就那个MNIST数据集，跑通一遍你就知道整个流程是咋回事了，那种成就感真的不一样哦。 Start with the simplest handwritten digit recognition, that MNIST dataset, get it running once and you'll know how the whole process works, that sense of accomplishment is really different.

你别想着一口吃成个胖子，上来就要做个混元大模型、豆包大模型、千问大模型、GLM、ChatGPT大模型，那不得把你吓死啊。 Don't think you can get fat in one bite, starting out wanting to make a big model like ChatGPT, that would just scare you to death.

慢慢来，先做小项目，做着做着你就懂更多了，不会的就去搜，现在网上教程一堆一堆的，啥问题搜不到啊。 Take it slow, start with small projects, as you keep working you'll understand more, if you don't know something just search it, there are tons of tutorials online now, what problem can't you find an answer to.

还有啊，数学真的不用怕，不是说你得先把高数线性代数全部考满分才能学深度学习，你边做边补就行，用到啥补啥，比你先傻坐着全学一遍有用多了。 Also, you really don't need to be scared of math, it's not like you have to get full marks in all of advanced math and linear algebra before you can learn deep learning, you can fill in the gaps as you go, learn what you need when you need it, it's way more useful than just sitting there learning everything first.

我见过好多人，上来先花三个月学数学，学完就忘了，也没动力继续学深度学习了，那不白忙活了吗。 I've seen so many people, start out spending three months learning math, forget it all by the time they're done, and don't have the motivation to keep learning deep learning, wasn't that all wasted effort?

做就行了，怕啥啊，做错了再改嘛，深度学习调参不就是试出来的吗，谁还不是一路踩坑踩过来的。 Just do it, what are you afraid of, fix it if you do it wrong, tuning parameters in deep learning is all about trial and error anyway, no one gets here without stepping on a bunch of pitfalls.

你可能会说，这都是大公司大专家玩的东西，跟我一个普通打工人有啥关系啊。 You might say, this is all stuff for big companies and big experts, what does it have to do with me an ordinary worker.

那你可就错大了啊，现在深度学习已经往各个行业钻了，以后不管你干哪行，说不定都能用上沾点边。 You're so wrong about that, deep learning has already seeped into every industry now, no matter what line of work you're in later, you might end up touching it somehow.

就说开饭馆的吧，现在都能用深度学习预测明天能卖多少份菜，准备多少食材不浪费，这不就是帮你省钱吗。 Take restaurant owners for example, they can now use deep learning to predict how many dishes they'll sell tomorrow, how much ingredients to prepare so you don't waste anything, that's just saving you money, right.

开水果店的，都能用AI帮你分拣水果，挑出坏果，比人挑得还快还准，省多少人工成本啊。 Fruit shop owners can use AI to sort fruit for you, pick out bad fruit, it's faster and more accurate than people picking, saves so much labor cost.

以后会有越来越多低门槛的深度学习工具出来，不用你会写复杂代码，拖拖拽拽就能用，普通人也能拿来解决自己的问题。 In the future there will be more and more low-threshold deep learning tools, you don't need to know how to write complicated code, just drag and drop and you can use it, ordinary people can also use it to solve their own problems.

你说未来会不会有那么一天，咱们普通老百姓，想做个自己的小AI，就像现在做个PPT一样简单？那可太爽了吧。 Do you think there will be a day in the future, when us ordinary people want to make our own little AI, it's as easy as making a PowerPoint now? That would be so cool, right.

当然了，现在深度学习也不是万能的，它也有毛病啊，比如有时候会瞎猜错，还需要好多数据才能训好，小公司玩不起大模型，这些都是问题。 Of course, deep learning isn't omnipotent now, it also has flaws, like sometimes it guesses wrong randomly, it also needs a lot of data to train well, small companies can't afford to play with big models, these are all problems.

但哪有技术一开始就是完美的啊，不都是慢慢改慢慢变好用的吗。 But what technology is perfect right from the start, isn't it all slowly improved and slowly gets better to use?

反正我觉得吧，多了解一点深度学习没啥坏处，哪怕你不做这行，知道现在AI到底是咋干活的，也不会被那些乱七八糟的AI噱头骗了对吧。 Anyway I think, knowing a little more about deep learning doesn't hurt, even if you don't work in this field, knowing how AI actually works now, you won't get scammed by all those messy AI gimmicks, right.

你要是感兴趣就赶紧动起来试试啊，说不定你就能用深度学习做出来个有意思的小东西，改变改变自己的小生活呢，你说是不是这个理。 If you're interested just get started and try it right now, maybe you can make an interesting little thing with deep learning, change your little life for the better, isn't that right.