Some ideas about AI development

  • 2021年6月4日
  • 2021年6月3日
  • lan

Over the past month, I've researched machine learning and deep learning to solve some agendas for AI data training.

At first, I realized that I had little or more misunderstand about AI. Before that the AI which I thought about was, something like facial detection or automatic driver or anything else like technology invented by famous corporations which are always come to my mind.

In fact, that is not right. For example, although computer vision is such a hot topic now. [ Computer vision provides machines with a human visual system to automate image collection, image processing, and image analysis. ]

But it is only a small part about AI. The human brain also has other abilities such as estimation ability, thinking ability, and motivation.

What we see, what we hear, what we speak, which are clearly felt by ourselves.

The AI researcher told us that something like “dark matter” is also included. What is the meaning of” dark matter”? It means something we cannot feel directly. Such as the computer vision, the image could be caught and viewed. But the other things can’t be seemed such as inference, consideration, motivation and so on which are most part of AI.So to find more truth of AI, We should act like Holmes to find clue of crime case.

Find out the secret of AI(cite from

a clever crow tries to take out the flesh from the nut by making use of the traffic environment of cross road, its action just tell what is “artificial” (cite from

When you stay in a room by yourself, you may never realize that how complex actions have been made by yourself. By turning viewpoint from here and there, walking to catch a cup of water, or trying to relax on sofa and so on. You may never image that how complex the action is that even nowadays the AI technology is not easy to imitate and analyze.

the robot imitates human’s action (cite from

In a word, AI should not be limited by what we know now. The more we explore it, we could invent the more useful AI production.

Next, I would like to talk about theory and programming. AI’s content is so big that we can hardly know most of it. Almost every day, around the world, so many AI related papers were upload on web. At the same time, large amount of technology companies highly competes around it to get business profit. For most of person, it is not an easy way to know about the fact of AI(except you are good at math or physics).As an AI engineer ,maybe catch the latest theory is not bad from the effective way which we could find most to solve the problem .Tuning the parameter of model is also important, but considering the input data’s feature engineering ,data preprocessing should cause our attention. On the other way, nowadays many tools could help programming and tuning to release them rapidly, so the programming should not be the core part of AI’s related job.

 A standard machine learning pipeline(cite from https://rstudio-pubs-

Feature engineering (cite from engineering)

Until now, our jobs are not so fresh because many similar of them have been done by others.

But I also think it's possible to learn great hints on tuning or data preprocessing with the knowledge and experience of other development engineers.  A lot of training data has been released, so I would like to make effective use of it.

Finally, I would like to mention a little about my challenges.  Currently, I am studying the detection and analysis of vehicle number plates.  At the same time, I am also learning about image processing, feature engineering, python, and deep learning.

 In the next blog, I will introduce the details about them.

 Thank you for reading.





実際にはこれは正しくありません。例えば、今コンピュータービジョンは確かに人気です。(コンピュータービジョンは、人間の視覚システムを機械に与え、画像収集・画像処理・画像解析を自動化するものです。) ただし、これはAIの極めて一部分です。人間の頭脳には、それ以外にも推定力や思考力、モチベーションといった能力が備わっています。それらは、直接に見ることはできませんが大きい割合を占めています。研究者は、それらを「ダークマター」と呼びます。このダークマターをどのように探し出したら良いでしょうか。いくつかの手がかりから、シャーロック・ホームズが謎解きをするようにAIの真実を探究することに意味があります。



次にAIの理論とプログラミングの話をしたいと思います。AIに関するコンテンツは非常に大きく、全てを把握するのはほぼ不可能です。ほぼ毎日、世界中で多くの AI 関連の論文が論文サイトにアップロードされています。同時に、多くのテック企業はビジネス利益を得るために激しく競争しています。

ほとんどの人にとって、AI を理解することは容易では無いでしょう。(数学や物理学が得意な場合を除きますが)。 AI エンジニアとしては、おそらく最新の理論を自らが見つけた効果的な方法から理解することは悪いアプローチではないと思います。また、AIモデルのパラメータを調整することも重要ですが、入力データの特徴量を考慮したデータの前処理に注意が必要です。 その一方で、最近ではプログラミングやチューニングを補助する多くのソフトウェアがあり、迅速に製品をリリースできるようになっているため、純粋なプログラミングは AIの中核となるべきではないでしょう。