What is AI, machine learning, and deep learning?

AI refers to concepts and technologies that artificially imitate human intelligence. The concept was proposed in the 1950s to allow computers to provide solutions to specific problems. However, it was not possible to find a solution without clearly defined rules, that is, without an algorithm created by humans.

Machine learning is a technology that enables machines to experience and learn from humans. In other words, a machine analyzes a large amount of input information (=data), finds rules and judgment criteria from that data, and makes predictions (inferences) about unknown data. However, having a computer perform the “finding of data features” step in the learning process is unstable and prone to errors, and as a result, it was necessary to manually set the conditions for feature extraction.

Deep learning (deep learning) is a deep neural network (artificial neuron) in which many layers of structures that imitate human nerve cells are superimposed, in which a machine automatically extracts the feature value of data without human intervention. It is learning using a network). In other words, deep learning has made it possible for computers to learn the feature extraction process of machine learning on their own.

In recent years, IoT (Internet of Things), which collects large amounts of data, and GPUs (Graphic Processor Units), which process that data with high computing power, have spread, and machine learning algorithms have evolved further. has accelerated the evolution of deep learning, and is said to have surpassed the human eye in the field of image recognition in recent years.

AIの定義と歴史を詳しく知りたいかたはこちらの記事をご覧ください。

https://lab.fujiele.co.jp/articles/4857/

How to utilize AI technology in business

An example of using machine learning is a recommendation system that combines purchase history and customer information, recommends products to customers who are likely to be interested, and encourages them to purchase. In addition, it can be used in the marketing field, such as a function that finds interests and preferences from data such as browsed articles and news and displays recommended articles.

ディープラーニングを活用した例では、製造現場での生産、組立、検査または設備の保守管理など、これまで人の五感に頼っていた工程を、ディープラーニングを搭載したロボットに置き換えることが可能となりました。製造現場での一つのタスク/工程での小さな改善は、年単位に換算すると非常に大きな利益につながります。その改善で得られた貴重な人的リソースは、より付加価値の高い工程へシフトする取り組みが世界でも広まっています。

また、顔認証技術も高度化し、人物の動画や遠く離れた人の写真からでも人の特徴を判別できるようになりました。マスクやサングラスをしていても指名手配犯を見つけ出したり、性別や年齢の属性を推測し、そのデータをマーケティングに活用したりと、あらゆる場面での活用が期待されています。

Start by setting goals for what you want to solve with AI

When using AI, it is important to set goals for what you want to achieve and what problems you want to solve with AI technology. It doesn't mean that "if you introduce AI, you can easily achieve this or that without the help of people". It is necessary to establish a path for how to solve specific business problems using AI technology.

The “AI model” is what specifically realizes “what you want AI to do”. There are several processes involved in creating an AI model, and specialized engineering skills are required.

Until the core AI model is created

Machine learning finds features from huge amounts of data and discovers rules for identifying them. This law is called a “learning model”. A learning model derives an output value from an input value, so to speak, it becomes the brain of AI. Generating a learning model involves performing calculations using a large amount of training data.

Also, this learning model is not completed once it is created. In order to improve accuracy, the data input to the learning model is also important. Prepare what kind of data and how many, adjust the parameters that require manual intervention, and repeat the creation of the learning model over and over. In addition, there are cases where only one learning model performs a task, but in reality, multiple learning models are combined to develop an "AI model" that solves the ultimate problem.

Only when an AI model is created in this way will it be possible to achieve what you want the AI to do.

Can AI introduction and AI development be started immediately?

AI introduction and development can be roughly divided into three phases: planning, verification/main development, and operation. As I explained earlier, the important tasks when starting the introduction and development are goal setting and preparation for AI model development.

Target setting can be advanced if there is a place to extract internal issues and discuss how to use AI technology in business. On the other hand, even if you want to start AI model development immediately, you can't start without data for learning, and you can't do anything without an environment for AI development. From here, I will introduce how to prepare the necessary environment for AI development, and how to prepare in advance.

Environment required for AI development

必要な環境は、ハードウェア、ソフトウェアと2つに分けられます。AI開発には膨大なデータを取り扱うため高速処理が必須となりますので、ハードウェアとしてはGPU（Graphic Processor Unit）と呼ばれる高速演算処理用のプロセッサーが必要となります。そのGPUは性能によって価格が大きく異なりますので、ご自身の環境にあうGPUを選定する必要があります。

一方ソフトウェアの開発環境としてはAIモデル開発に欠かせないフレームワークやライブラリーを入手したり、昨今は環境が一式簡単に揃うコンテナ技術を活用した開発が盛んになっており、この辺りの専門知識を習得し、環境を構築しておく必要があります。

In order to avoid the situation where you want to proceed with AI development, but the development environment becomes a threshold and you can't start development for a long time, it is a wise choice to use a service that prepares the necessary environment according to each person. increase.

In recent years, the development environment has increased various options according to requests and applications, such as utilizing the cloud and housing services. However, I think that companies that have an idea of the following examples are considering preparing an environment for AI development on their own (on-premises).

■学習用データを外部に開示できない

■学習させる度にデータを移動させると開発効率が悪い

■クラウド利用に制限がかかっている

Here we go! Get started with AI development

本記事ではAI導入・AI開発を始める前に知っていただきたいことをご紹介しましたがいかがでしたでしょうか。

AIをビジネスに活用したいと感じられて、今すぐ自社のエンジニアに始めさせたい、自分たちで始めたいという方にとって、環境構築がボトルネックになることがあります。続く記事では初心者向けのAI開発環境の作り方をご紹介します。

Contact Us