The winter break has begun, and Science Popularization China has specially launched a series of popular science content about technology and Artificial Intelligence (AI), hoping to introduce the mysteries of these cutting-edge technologies in an easy-to-understand manner.
From basic concepts to practical applications, from historical development to future trends, children can appreciate the charm of technology here and learn how technology shapes our world.
Welcome both big kids and little kids to watch together!
Artificial intelligence has infiltrated every aspect of our lives. Nowadays, most of the artificial intelligence products we use are inseparable from an important technology—artificial neural networks. Interestingly, artificial neural networks were not well-regarded at their inception and faced a cold reception for two to three decades.Today, let’s talk about the story behind artificial neural networks.
The Birth and Challenges of Artificial Neural Networks
When computers were first born, people had many beautiful visions for them, hoping they could replace human thinking and become humanity’s “second brain”.
At that time, there were two main ideas.
One idea was that human reasoning, such as causality, syllogism, and inductive reasoning, could all be expressed in corresponding mathematical symbols.Therefore, as long as we let computers master the rules behind these symbolic reasoning, they could simulate human thinking.This idea represents the symbolic school in the field of artificial intelligence, emphasizing knowledge bases and logical reasoning.
The other idea was that we need to simulate the basic structure of the human brain to mimic human thinking. This is the main viewpoint of another important school in artificial intelligence, known as the connectionist school.
The human brain is actually composed of numerous neurons. Each neuron performs simple operations, such as receiving information, processing it, and outputting new information.Simulating a neuron is not difficult; thus, if we establish numerous neurons in a computer, forming a network similar to the brain, we could simulate complex human thinking, right?
Image copyright; using it may lead to copyright disputes
Following this line of thought, in 1946, scientists Warren McCulloch and Walter Pitts proposed the concept of “artificial neural networks.” In 1956, the first artificial neuron was born. This neuron is called a perceptron, which can give simple judgments based on input information.
If we judge whether an apple is of good quality or not, we will consider several factors: size, color, aroma, and sweetness.
By inputting scores for these aspects, the perceptron can make a judgment based on the scores. Of course, the perceptron can assign different weights to different features; for example, size is not so important, so it is given a weight of 0.4; color is more important, given a weight of 0.6; aroma and sweetness are both very important, assigned weights of 0.8 and 0.9, respectively. After calculation, the perceptron can quickly determine the quality of the apple.
However, this perceptron model is too simple; it can only perform basic classifications, and people at the time did not think it could help computers achieve any real thinking, leading to a winter for artificial neural networks.
During this period, people tried to increase the number of layers in the neural network. For example, in the previous apple evaluation example, we could add several hidden layers, with some hidden layers focusing more on appearance, increasing the weights of size and color, while others emphasized taste, giving higher weights to aroma and sweetness. People hoped that this approach would enable the neural network to make better judgments.
But this still did not change the fate of neural networks being neglected until the 1970s and 1980s when the backpropagation algorithm emerged, leading to a significant advancement in artificial neural networks.
Breakthrough of the Backpropagation Algorithm
What is the backpropagation algorithm? Let’s continue with the apple classification example.
Suppose we input several parameters of a certain apple, and the neural network outputs “good apple.” However, in reality, we consider this to be a “bad apple.” Through feedback, we tell the neural network the correct result, and the neural network will reflect on itself, adjusting the weights until it outputs the correct result. We call the process of “backtracking and adjusting weights” backpropagation.
With the backpropagation algorithm, artificial neural networks can continuously self-adjust and optimize to reach more reliable conclusions. Combined with the previous step of adding hidden layers, neural networks can handle very complex problems.
This has led to the various conveniences we take for granted. For example: mobile apps learn from our historical data to find videos we might be interested in; autonomous driving recognizes where the road is and where pedestrians are based on extensive image training; voice assistants use neural networks to understand the intent behind what people say.
Technological development is often unexpected; no one could have imagined that a technology that had been neglected for decades could be revived and play such an important role in our lives.
In the next episode, we will discuss convolutional neural networks. What new heights will this groundbreaking innovation take neural networks to?
Planning and Production
This article is a product of the Science Popularization China – Creative Cultivation Program
Produced by | Science Popularization Department of the China Association for Science and Technology
Supervised by | China Science and Technology Press Co., Ltd., Beijing Zhongke Xinghe Cultural Media Co., Ltd.
Author | Beijing Yunyujin Cultural Communication Co., Ltd.
Reviewed by | Qin ZengchangBeijing University of Aeronautics and AstronauticsSchool of Automation Science and Electrical EngineeringAssociate Professor
Planned by | Fu Sijia
Edited by | Fu Sijia
Related Recommendations
1. New Research! Exercise Can Alleviate Anxiety and Depression! The Best Effects Are from These 3 Types
2. The Warmest Sleeping Tips in Winter! Super Simple, But Many People Don’t Know
3. What Differences Will Occur Over Time Between People Who Often Open Windows for Ventilation and Those Who Don’t?
4. Cold-sensitive People vs. Cold-resistant People, Which Is Healthier? The Truth Is…
5. Can You Find Cultural Artifacts in Green Belts?
The cover image and images in this article are from copyright libraries
Reprinting may lead to copyright disputes. For original text and images, please reply “Reprint” in the background
Light Up “Looking”
Let’s Gain Knowledge Together!
