The Artificial Intelligence explained to my grandmother

Artificial Intelligence (AI) is an area of computer science that aims to create intelligent machines. There are many applications for AI such as knowledge, reasoning, problem-solving, perception, machine learning, planning, the ability to manipulate and move objects, and so on. In fact, machine learning is a very important part of Artificial Intelligence.

Machine learning consists of giving data or images to the machine-learning algorithm (such as supper vector machine (SVM) algorithm) to train it to do something. For example, you can feed the machine learning system with cat and dog photos. So, the machine learning system learns the cat and dog’s features using the SVM algorithm. Then, the machine learning can classify which photos have a dog or a cat. The machine learning can use the training images or new images to classify the photos.

Another example of machine learning is movie recommendations.

So, machine learning is used to give movie recommendations based on user’s past watching history and other users’ watching history.

Nowadays, deep learning, which is part of machine learning, is very popular. Deep learning uses artificial neural network (ANN) algorithm instead of machine learning algorithms such as SVM algorithm. In fact, deep learning is more accurate than traditional machine learning algorithms. Furthermore, deep learning has many applications such as chatbots, object recognition, classification, self-driving cars, deep art, and so on.

The idea of Artificial Intelligence is to create a machine that thinks, work and react like a human.

Machines can’t think like humans do, but they can beat human players.

For example, Deep Mind’s AlphaGo is an AI system (which uses artificial neural network algorithm) that was able to beat some of the best Go players in the world. Basically, it was used many game moves to train the AlphaGo. Hence, AlphaGo learned how to play the Go game. In fact, the new version of AlphaGo, called AlphaGo Zero, learned by trying out moves to see if they worked. It rediscovered 3,000 years of human knowledge of the Go game in just 72 hours and was able to beat the Go grandmaster Lee Sedol in three days.

Machine learning

In machine learning, there are two types of learning: supervised and unsupervised learning.

Machine learning: Supervised learning

In supervised learning, a machine learning system is trained to predict the “label” or “class” of the data and process accordingly. For example, if the machine learning system is fed with a lot of labeled images of animals, it will be trained to classify these images using the labels (which are the animal types in this example). Hence, the machine learning system predicts the “class” or “label” of unlabeled images. This means if you fed the system with a dog photo then the machine learning system must predict that it is a dog. So, the machine learning system finds general features in order to predict its “class”.

In fact, supervised learning is fast and demands less computational power than other training techniques because it requires training data what are examples of what you want it learns.

The most popular supervised machine-learning algorithm is support vector machines (SVM) but there are others : Multiclass SVM, Transductive SVM, Bayesian SVM, Support vector regression…

Machine learning: Unsupervised learning

In unsupervised learning, the data is not labeled. So, the machine learning system must find patterns in the data and classify it. For example, the machine learning system is fed with images of dogs and cats that are not labeled, so the machine learning system must find that there are two different types of animals.

In the case of unsupervised learning, it must be used an algorithm to find the class of the data. So, the algorithm must find natural clustering of the data to groups or “classes”, and the new data is mapped to these “classes”.  This algorithm is called support vector clustering, which applies the same statistics of support vector machine algorithm to categorize unlabeled data.

Deep learning

Deep learning is a very important part of artificial intelligence. Geoffrey Hinton and Andrew Ng are the pioneers that used an artificial neural network (ANN) for image recognition. So, they basically feed the deep learning system with a lot of images, and the artificial neural network was trained to find patterns of these images. Hence, the deep learning system learned how to recognize an object (such as a cat and a dog) from those images.

Artificial neural network (ANN) is an algorithm that imitates the human brain and has its own artificial neurons like the human brain. These artificial neurons are connected to each other and has some layers, and each layer has artificial neurons.

There are three types of layers: input, hidden, and output layer.

How it works ? The input layer receives the input data such as images, data or audio. This data and images are passing it through these layers that perform simple computations until the output layer give us the « class » of the object in the photo. So, using this algorithm and the input data, the deep learning system is able to recognize general patterns to predict the class or predict a value.

Deep learning neural network

Fig 1. Deep learning neural network

Basically, each input neuron receives a part of the data such as a part of an image. In fact, an image is a matrix that has the numerical value of each pixel. So, this matrix is the input data. Then, this data goes through the layers in order to find patterns. Then, the output layer predicts its “class”.

ConvNets

In the case of image classification, it is used a special type of neural network, which is called convolutional neural network or ConvNets. The convolutional neural network also has its input layer, hidden layers, and output layer. But in this case, the hidden layers are: convolutional layer, rectified-linear layer (ReLu layer), pooling layer, flatten layer, and fully connected layer.

It’s a little bit complicated to develop Grandma so let’s move to a real revolution : Deep learning !

Deep learning applications

Deep learning can be used for self-driving cars.

In this case, the deep learning system receives as input a video in real-time, and the AI system must predict the objects that are in front of the car. Furthermore, the AI system uses an algorithm to predict the trajectory of the car.

Another example of deep learning is that it can be used to diagnose patients. In this case, the deep learning system is trained using the input data such as x-ray images, and the AI system must diagnose patients. You can read my article on how artificial intelligence helps prevent cancer.

A popular deep learning application is chatbot.

These chatbots are deep learning system that learned how to « talk » with humans. So, these chatbots are able to translate and interpret human language input. The chatbots use a combination of natural language processing and deep learning.  Basically, these chatbots are trained using data as books, human chats, and  are able to have a chat with a human, write a book, translate from one language to another, and so on.

Another interesting deep learning application is deep art.

In this case, it is used deep learning to learn the style of an artist. So, this deep learning system receives as input data images as paintings or something of an artist in order to learn his/her style. Then, you can give a photo or an image and the deep learning system generates a completely new image. So, your own photos can be changed to have Picasso’s style.

Fig. 3 shows how deep learning was used to combine a photo with the Van Gogh’s style.

Deep art Deep learning

Fig 3. Deep art

Deep learning systems are able to learn from examples, data, images, audio, and so on. Machines are able to learn from their mistakes, make a prediction, take action based on that prediction, and learn from that action. This is called reinforcement learning. This type of learning becomes more accurate over time. In fact, reinforcement learning is applied in many applications such as games and self-driving cars.

Conclusion

In conclusion, artificial intelligence is very popular nowadays because there are many applications with machine and deep learning. From the simplest (movie recommendations) to the most complex (self-driving) artificial intelligence has already begun to change our lives. But don’t worry Grandma, no artificial intelligence can replace the love i have for you !

Par | 2018-04-09T02:19:24+00:00 dimanche, 8 avril, 2018|Catégories : E-Business, E-commerce, Expérience Client, Objets connectés / IOT, Tech|

À propos de l'auteur :

Curieux, je m'intéresse aux nouvelles technologies et à l'innovation. Contactez-moi => julienherry@yahoo.com #MarketingDigital #Chatbot #Robots #IA #IntelligenceArtificielle #Blockchain #IoT

Laisser un commentaire