Comic Edition: What is Artificial Intelligence? What is Machine learning? What is Deep Learning? Comic Edition: What is a Neural network?

This is a fun and simple introduction to deep learning, including its applications, how it differs from machine learning and artificial intelligence, what neural networks are, and how to train them to recognize handwritten numbers. Finally, some popular applications of deep learning are introduced.

How does Google translate an entire web page into different languages in just a few seconds, or how does your phone gallery group images according to their location? These are the results of deep learning.

Deep learning is a subset of machine learning, which in turn is a subset of artificial intelligence.

Artificial intelligence is a technology that enables machines to mimic human behavior, machine learning is a technology that implements AI by using algorithms trained by data, and finally deep learning is machine learning inspired by the structure of the human brain (biological neural networks). Such structures are known in deep learning as artificial neural networks.

Let’s better understand deep learning and how it differs from machine learning.

Suppose we have a machine that can distinguish between a tomato and a cherry. If we use machine learning to do so, we must tell the machine what features it can distinguish between the two. These traits could be Size and Type of Stem.

Using deep learning, neural networks can extract features without human intervention.

Of course, this feature requires having a lot of data to train our machines. Now let’s delve into how neural networks work.

The principle of

Here, we have three students who each write the number 9 on a piece of paper, but they don’t write exactly the same. The human brain can easily recognize numbers, but how does a computer recognize them? This can be done using deep learning.

Each number is displayed as a 28×28 pixel image for a total of 784 pixels.

This is a trained neural network to recognize the handwritten Numbers, neurons are the core of one of the most basic neural network entities, is a place where information processing, 784 pixels in each pixel is delivered to a neural network, the first layer of neurons that formed the input layer, input layer only accept input, not to manipulate the function.

The other end is the output layer, and the layer of neurons between the input layer and the output layer is called the hidden layer. The neurons of hidden layer and output layer are functional neurons with activation function.

The input layer neurons receive 784 inputs, which are transmitted through weighted connections, and the total input values received by the neurons are compared with the neuron’s threshold (each neuron has a threshold called Bias), The output of the neuron is then processed by Activation Function, the result of which determines whether the neuron is activated or not.

The learning process of neural network is to adjust the “connection weight” between neurons and the threshold value of each functional neuron according to the training data. In other words, what neural networks learn is embedded in connection weights and thresholds.

application

So what are the applications of deep learning?

In customer service: When most people talk to customer service agents, the conversation seems so real that they don’t even realize it’s actually a robot.

In the medical industry, neural networks can detect cancer cells and analyze MRI images to provide detailed results.

Self-driving cars, like science fiction, are now a reality. Companies like Apple, Tesla and Nissan are working on self-driving cars.

limitations

Deep learning is broad in scope, but it also faces some limitations.

The amount of data

The first limitation is data. While deep learning is the most efficient way to process unstructured data, neural networks require large amounts of data to train.

Ability to calculate

Let’s say we have a lot of sample data, but not every machine has the capacity to process it, which brings us to the second limitation: computing power. It is often shortened to “arithmetic power”.

Training neural networks requires thousands of graphics processing units. Gpus are certainly more expensive than cpus.

Training time

Finally, there is the training time. Deep neural networks need hours or even months of training, and the time increases with the increase of data volume and layers in the network.

Deep learning framework

Some popular deep learning frameworks include Tensorflow, Pytorch, Caffe, DL4J, and Microsoft Cognitive Toolkit.

In the future

We’ve only just scratched the surface of what deep learning and AI can do in the future, and the future will be full of surprises.

Horse Technology is developing a device for blind people that uses deep learning to describe the world to the user using computer vision, replicating the human mind as a whole.

quiz

So here’s a quiz on the correct order in which neural networks work:

  • A. The bias is added B. The bias is added
  • B. The weighted sum of The inputs is calculated
  • C. Specific neuron is activated
  • D. The result is fed to an activation function

The answer:

  • B. Calculate the weighted sum of the inputs
  • A, plus deviation
  • D. The result is entered into the activation function
  • C. Specific neurons are activated

Description: In a neural network, each neuron in a layer is connected to other neurons in the corresponding layer. These connections have random weights. Computes the weighted sum of the inputs and adds an additional input in bias form (w * x + b). The results are entered into the activation function. Based on a certain threshold, only those neurons above that threshold will fire.