Guide To Understand The Machine Learning, Deep Learning, AI

Introduce, this is Samantha. She is your personal assistant who came from 2025. He can sort emails , arrange planning meetings, even order your daily needs. She can also paint or compose poetry. She is your best friend. It is an artificial intelligence (AI) from the film, which illustrates how Siri, with abilities that have been far more enhanced, can change human life.

At present, large and small technology companies are competing to make this happen. You also must have read the news about this. You must have heard the jargon such as AI, machine learning ( machine learning ), deep learning , neural networks ( neural networks ), or natural language processing ( natural language processing ).

It may sound a little confusing. Therefore, let us consider the basic explanation of this concept and its relevance.

Understanding Machine Learning, Deep Learning,and AI

What is Artificial Intelligence or AI?

Simply put, AI is a way to make computers think intelligently or beyond human intelligence. The aim is that computers can have the ability to behave, think, and make decisions like humans.

There are two kinds of artificial intelligence:

AI with limited ability, or weak AI

This type of AI is designed to complete simple tasks. These AIs are already around us and they are even able to defeat humans in chess games. Digital assistants like Siri and Cortana are able to give us news about the weather. In addition, automatic control cars have also been milling about on the streets.

[Related Article: Artificial Intelligence Trends Forecast ]

However, their abilities are limited. Automatic control cars cannot be invited to play chess. Siri can’t read and delete emails you don’t need. AI’s ability at this level is very limited and is unable to do anything outside its original program.



AI with unlimited abilities or strong AI

From this point we begin to enter the realm of science fiction. Samantha is the most appropriate definition to describe this. He can learn new things and modify his own code base. He can beat you in chess and can be your personal driver.

Anatomi Artificial Intelligence

Now you understand that AI with unlimited abilities is the ultimate goal. How do we get there? There are five things that AI needs to master:

1. Perception: Like humans, a computer needs the five senses to interact with the world. However, the number of computers can be more than five. Computers can be equipped with senses that humans do not have. Exceptional vision and hearing? Everything can be realized through the help of a machine .

2. Natural language processing (NLP) : In addition to sensing capabilities, AI must also be able to convey language verbally or in writing. They need to be equipped with the ability to identify sentences and understand their differences, accents, and meanings. This is a very difficult task for the machine, remembering the same sentence can mean different things depending on the context.

[Related Article: Python Interpreter Tutorial]

3. Delivering knowledge : After AI is able to perceive various things β€” objects, humans, concepts, words, and mathematical symbols β€” AI must find ways to convey all information in the world through its own thinking.

4. Decision making : After AI collects data through its senses and connects existing concepts, AI can use these data to solve problems logically. For example, chess game software can identify movements carried out by human players and then launch its own strategy.

[Related Article: Programming Languages]

5. Planning and mapping : To be more human, AI cannot only think like humans. He must be present among us. Therefore, researchers are looking for ways to help AI map the three-dimensional world and plan the most effective routes.

The ability of an automatic control car must clearly be increased, because one mistake can endanger human life .

The Singaporean prime minister is trying out an automatic control car. Image source: Kenji Soon, MCI

You can find linkages between these five things in certain areas such as machine vision , namely the fields used in conducting imaging and analysis to solve problems.

For example, Facebook, which studies the photos that you upload on their social media to suggest who you should tag . Uniquely the results can be accurate .

Automated control cars are perhaps the most complex machine vision implementation at the moment. He must be able to recognize traffic signs, observe traffic conditions, and pay attention to the presence of humans, objects, and other cars. He also must continue to function in weather conditions with even the worst visibility, day and night, and on roads that are decent or unfit to pass.

The things needed to get there

Actually this is not all really new concepts. The concept above was presented at the beginning of 1956 at the Dartmouth conference which is often claimed as a milestone in the field of information in the field of AI.

Although it will take decades for technology to be in line with the human imagination, in the end we may be on the verge of AI revolution , with VC investment increasingly abundant, the proliferation of leading technology companies involved in research and development, and increasingly the many uses of AI in our lives.

One of the valid factors contributing to the development of AI is Moore’s Law, which allows the creation of microprocessors with greater computational capacity in smaller sizes. Computational capabilities have reached the point where AI has functioned well and the price is also increasingly affordable.

Big data is another area that plays a role in the rise of AI: Google made a breakthrough in 2012 when they created a neural network that was supplied with very large data and consisted of 10 million YouTube videos randomly.

The result?

The neural network is able to study the appearance of cats without being taught by humans. The level of accuracy in identifying these furry animals reaches 75 percent.

When the machine is able to learn

Now we try to explain some concepts that are often confused with other concepts.

Machine Learning

Machine learning is an AI technique that is related to learning data and using it to predict information in the world.

Machine Learning

Image source: Wikipedia

Machine learning is built using algorithms.

[Related Article: Machine Learning Interview Questions]

This series of instructions will solve a problem. An example of the algorithm in question is a decision tree learning and association rule learning .

However, the machine learning algorithm that plays a role in life in the world is artificial neural networks, a technique inspired by the workings of neurons in the human brain.

Even this technique has penetrated pop culture: in the Silicon Valley comedy series , there is a startup called Pied Piper that runs compression services on neural networks.

Simply put: a neural network consists of several layers of neurons. Inputenters through the first layer. Each neuron receives input , so each neuron has a charge, and produces output based on their charge.

The output of the first layer is then distributed to the second layer for processing, and so on until the final output can be produced.

Then interesting things happened.

Anyone who runs the network can define what the “right” final output should be. Every time the data is distributed through the network, the end result is compared with the “right” results, and a number of improvements will be made until the correct final output is created . In other words, the network is able to train itself.

This artificial brain can learn how to identify many things. For example, a chair in a photo. Over time, he can learn the characteristics of the chair, and improve his ability to identify the object.

AI’s Facebook director, Yann LeCun, explained neural networks through an analogy :

If the pattern recognition system is likened to a black box with a camera on one end, a green light and a red light at the top, and various buttons on the front, the learning algorithm will try to adjust the button under certain conditions. Say a dog passes in front of the camera, the red light is on, and when it passes is a car, a green light that lights up.

You show a dog on the machine. If the red light is bright, don’t do anything. If the red light is dim, adjust the button so that the light becomes bright. If the green light is on, set the button so that the light dims. Then show a car, and set the button so that the red light dims and the light of the green light gets brighter.

[Related Article: VueJS Tutorial]

If you show a variety of cars and dogs, and you keep setting the buttons little by little all the time, in the end the machine will understand the correct answer.

Machine learning is built using algorithms

Image source: a16z

Deep Learning

Now we come to the discussion of deep learning , which can be interpreted as a series of methods for training multi-layer artificial neural networks.

Apparently, this method is effective in identifying patterns from data. When the media talks about neural networks, the possibility in question is deep learning.

See reviews on machine learning and deep learning in this video:

Deep learning has significantly affected the progress of AI development . Not only software, but its use has penetrated various other industries.

Facebook also uses deep learning in M , an AI-based virtual assistant that helps users complete their tasks such as doing research, ordering flight seats, and ordering coffee.

[Related Article: What is Big Data]

Google uses a deep learning system called RankBrain to filter search results. This goes hand in hand with a number of conventional methods, as described by Bloomberg :

This system helps Google handle 15 percent of questions per day that this system has never received before. For example, this system adjusts itself when faced with ambiguous questions such as, ‘What are the names for consumers at the highest level of the food chain?’

This system is increasingly useful and is now in the position of the top three Google search results factors, outside of links and content.

Is the system capable of identifying if the cat is seen? That is deep learning .

[Related Article: Database]

From Siri to Samantha

Deep learning may be the main puzzle piece that can bring people to the creation of a more intelligent and humanized AI.

The Google brain that is capable of scanning cat images requires 16 thousand computer processors to run. AlphaGo, the program that managed to defeat Go champion , Lee Sedol, runs “only” with 48 processors. Someday, it is not impossible for neural networks to work on  low-priced smartphones.

Deep learning can enhance all parts of AI, from natural language processing to machine vision . Think of deep learning as a better brain that can improve the way you learn computers.

He can improve the ability of virtual assistants such as Siri or Google Now to handle things that have not been well recognized by the two virtual assistants. Video processing and clip making are also very possible to do by deep learning.

Who knows, someday deep learning can win an Oscar.

Leave a Comment