Besides "artificial intelligence," we often hear terms like "machine learning" and "deep learning"... What do these terms mean? What's the relationship between them? Let's find out together!
Artificial Intelligence
When it comes to artificial intelligence, the first thing that comes to mind is probably the robots with human-like intelligence in science fiction movies. But in reality, artificial intelligence is more than just robots.
Artificial intelligence (AI) was proposed by John McCarthy in 1956, initially defined as "the science and engineering of making intelligent machines." Today, AI refers to "a new technical science that studies, develops, and applies theories, methods, technologies, and application systems to simulate, extend, and expand human intelligence." Sounds a bit convoluted, right? Let me summarize: AI aims to enable machines to simulate human thinking abilities, allowing them to perceive, think, and even make decisions like humans. Today, AI is no longer a single discipline but an interdisciplinary field involving computer science, psychology, linguistics, logic, philosophy, and many other disciplines.
Artificial intelligence may seem like a sophisticated technology, but it's actually a very broad concept. Various forms of AI already exist around us, such as autonomous driving, facial recognition, intelligent robots, and machine translation.
Faced with a wide variety of artificial intelligence, we can categorize them into three types based on their capabilities:
Artificial Narrow Intelligence (ANI)
Artificial intelligence that excels in a particular area can only perform specific tasks. For example, a facial recognition system can only recognize images; if you ask it what the weather will be like tomorrow, it won't know how to answer.
Artificial General Intelligence (AGI)
Human-level artificial intelligence is capable of exhibiting intelligence similar to humans in multiple fields, understanding, learning, and performing various tasks. Currently, strong AI has not yet been realized and remains a long-term goal of AI research.
Artificial Superintelligence (ASI)
Artificial intelligence that surpasses human intelligence is smarter than humans in every field, capable of performing any intellectual task and outperforming humans in many ways. Although superintelligence frequently appears in science fiction, it remains a theoretical concept in reality and is not yet a possibility of being realized.
Speaking of which, I'd like to ask everyone, what kind of artificial intelligence is AlphaGo, which defeated the world Go champion?
Although AlphaGo defeated humans in Go, its ability was limited to Go. It couldn't play Gomoku or Ludo, so AlphaGo is considered a weak artificial intelligence.
Machine Learning
As mentioned earlier, the goal of artificial intelligence is to enable machines to think and make decisions like humans. But how exactly can this be achieved?
Think back to when we were born, we knew practically nothing. After decades of learning, we acquired various kinds of knowledge and skills. Machines are the same. To enable them to think, they must first learn, summarize patterns from experience, and then possess a certain degree of decision-making and discernment abilities. This is the core of artificial intelligence—machine learning.
Machine learning specifically studies how computers can simulate or implement human learning behavior, acquire new knowledge and skills through learning, thereby reorganizing existing knowledge structures and continuously improving their own performance.
Machine learning is a multidisciplinary field that involves probability theory, statistics, approximation theory, algorithm complexity theory, and many other disciplines.
How do machines learn? Let's first look at the human learning process:
Attending class: learning theoretical knowledge and absorbing knowledge.
Summary and Review: Reinforce understanding through review.
Organizing a knowledge framework: Organizing knowledge and forming a system.
Homework: Through practice, further deepen your understanding.
Weekly quizzes: to check understanding
Identify and fill gaps in knowledge: Improve learning methods
Final exam: to check final learning outcomes
The machine learning process is similar, including the following 7 steps:
Data Acquisition: Collecting relevant data
Data processing: Transforming data and standardizing its format.
Model selection: Choosing a suitable algorithm
Model training: Using data to train the model and optimize the algorithm.
Model evaluation: Evaluate model performance based on prediction results.
Model tuning: Adjusting model parameters to optimize model performance.
Model prediction: Predicting unknown outcome data
In short, machine learning is the process of automatically extracting logic or rules from data using algorithms, and then making predictions based on the results of the extraction and new data.
For example, if we want a computer to recognize a dog when it sees it, we need to show it a large number of pictures of dogs and tell it that this is a dog. Through extensive training, the computer will summarize certain patterns, and when it sees a dog again, it will capture the corresponding features and conclude "this is a dog." If the algorithm is not perfect, it might mistake a cat for a dog. Therefore, the computer needs to automatically improve the algorithm based on empirical data to enhance its predictive ability.
Based on the learning method, machine learning can be divided into the following four categories:
Supervised learning
Learning from labeled data, which contains independent and dependent variables, involves making predictions by learning from known input and output data, such as in classification and regression tasks.
Classification tasks : predicting the category to which data belongs, such as spam detection, identifying plant and animal categories, etc.
Regression task : Predicting data based on previously observed data, such as predicting house prices, height and weight, etc.
Unsupervised learning
Analyze unlabeled data, i.e., data with only independent variables and no dependent variables, to discover patterns in the data, such as through clustering and dimensionality reduction.
Clustering : grouping similar things together without focusing on what those things are, such as customer grouping.
Dimensionality reduction : By extracting features, high-dimensional data is compressed and represented in low dimensions, such as merging a car's mileage and years of use into a wear value.
semi-supervised learning
The training data is only partially labeled. Unsupervised learning is first used to process the data, and then supervised learning is used to train and predict the model. For example, a mobile phone can recognize photos of the same person (unsupervised learning). Once photos of the same person are labeled, newly added photos of that person will also be automatically labeled with the corresponding tags (supervised learning).
reinforcement learning
By interacting with the environment, the algorithm is optimized based on rewards or penalties until the maximum reward is obtained, thus generating an optimal strategy. For example, when a robot vacuum cleaner encounters an obstacle, it will optimize its cleaning path.
Deep Learning
Having learned the above, you should now be familiar with machine learning. So what is deep learning? And how is it related to machine learning?
Deep learning is a new research direction in the field of machine learning. It is an algorithm that uses multi-layered neural networks to learn and understand complex data. Machines learn complex tasks by learning deep representations of sample data, and eventually, they can have analytical and learning abilities like humans, such as recognizing text, images, and sounds.
Unlike traditional machine learning, deep learning uses neural network structures. The length of a neural network is called the "depth" of the model, hence learning based on neural networks is called "deep learning." Neural networks simulate the neuronal network of the human brain, where neurons can process and transform data. Through multi-layered neural networks, the features of data can be continuously extracted and abstracted, enabling machines to better solve various problems.
Typical deep learning algorithms can be categorized into four types:
Convolutional Neural Networks (CNNs) are commonly used for image recognition and classification tasks.
Recurrent Neural Networks (RNNs) are suitable for processing sequential data, such as in natural language processing.
Long Short-Term Memory (LSTM): A special type of RNN structure that is better able to handle long sequence data.
Generative Adversarial Networks (GANs) are used to generate new data, such as images, audio, or text.
With the support of deep learning, artificial intelligence has developed rapidly, and we believe that in the near future, we will have a brand new AI era.
Conclusion
Another piece of useful knowledge has been added. Let me briefly summarize it:
"Artificial intelligence" is a broad concept that aims to enable machines to think and perform tasks like humans.
Machine learning is a method for achieving artificial intelligence, with the aim of learning patterns from data. Traditional machine learning requires manual determination of data features.
"Deep learning" is a specific branch of machine learning that is based on neural networks and can automatically learn data features.
