The term "artificial intelligence" has been floating for some time. We see this in science fiction movies, the "AI" gaming robots we fight against, Google searches, and oh yes, those robots that will one day occupy the world. However, "machine learning" and "deep learning" have surfaced, and many people have asked what they are.
artificial intelligence
Artificial intelligence is a general category, and it is general to all three categories. In the figure, artificial intelligence will be a larger encapsulation circle that includes machine learning and deep learning. Artificial intelligence is basically any intelligence displayed by a machine, which can guide it to the optimal or sub-optimal solution for a certain problem. The simplest AI example can be found in the form of the Tic-Tac-Toe AI player. If the robot follows the following pre-programmed algorithm, it will never lose the game: (provided by Wikipedia)
1. If someone is "threatened" (ie twice in a row), take the remaining squares.
2. If a move “fork” causes two threats at a time, proceed with that move. Otherwise,
3. If the central square is free, take it. Otherwise,
4. If your opponent is playing in the corner, go to the opposite corner. Otherwise,
5. If there is an empty corner. Otherwise,
6. Take any empty square.
Now, algorithms like this do not have the ability to recognize, learn or solve problems that most people associate with "AI". However, an algorithm is only a proxy for the optimal solution for a given problem and its state.
Wikipedia provides another definition for artificial intelligence agents:
Any device that can perceive its environment and take action to maximize its success in achieving its goals
Agents that belong to AI but not machine learning are usually agents that only use decision trees for logic, or agents that are built using rules and instructions.
The 3 most popular Ai articles:
1. A beginner's guide to implementing RNN-LSTM using Tensorflow
2. Keras cheat sheet: Neural network in Python
3. Making a simple neural network
Machine learning: signs of cognition
Arthur Samuel coined the term "machine learning" in 1959 and defined it as "the ability to learn without explicit programming." The most basic form of machine learning is the practice of using algorithms to parse data, learn from it, and then determine or predict things in the world. For starters, the most common example is house prices. How do websites like Redfin or Zillow predict the current price of owning a house?
It's not that complicated. The essence of machine learning is that it actually does the best in many ways. The house price prediction model looks at a large amount of data, and each data point has multiple dimensions, such as size, number of bedrooms, number of bathrooms, yard space, etc. It creates a function based on these input parameters, and then moves the coefficients to each of these parameters to view more and more data.
This machine learning method is called "supervised learning", where the data provided to the model includes the answers to questions for each input set. It basically provides the input parameters (called features) and the output of each set of features, from which the model adjusts its function to match the data. Then, when given any other input data, the model can perform the same function and provide accurate output.
Other factions of machine learning are unsupervised learning and reinforcement learning. In short, unsupervised learning is just to find similarities in the data-in our house example, the data will not include house prices (the data is only an input, not an output), and the model can say "Well, Based on these parameters, house 1 is most similar to house 3" or similar, but the price of a given house cannot be predicted.
It is best to use simple, short diagrams to explain reinforcement learning: the
agent performs an operation in the environment, which is interpreted as a reward and state representation, and these operations will be fed back to the agent. Think of a little baby: crying will produce candy, which is reward. Over time, if parents continue to satisfy their children’s cravings for sweets, babies will learn to cry every time they want sweets.
Deep learning: connection with humans
Deep learning is inspired by the structure and function of the brain, that is, the interconnection of many neurons. Neural networks are algorithms that mimic the biological structure of the brain.
News from MIT:
Loosely modeled on the human brain, neural networks consist of thousands or even millions of simple processing nodes that are tightly interconnected. Most neural networks today are organized into layers of nodes. They are "feedforward", which means that data passes through them in only one direction. A single node may be connected to multiple nodes at the next level (receiving data from that node), and multiple nodes at the next level above it (which node sends data to).
A node will assign a number to each of its incoming connections, called a "weight". When the network is active, the node receives different data items-different numbers through each of its connections, and multiplies it by the associated weight. Then, add the resulting products together to get a number. If the number is lower than the threshold, the node will not pass any data to the next layer. If the number exceeds the threshold, the node "triggers", which in today's neural networks usually means sending the number (the sum of the weighted inputs) along all its outgoing connections.
When training a neural network, its weights and thresholds are initially set to random values. The training data is fed to the bottom layer (input layer), passes through the subsequent layers, and is multiplied and added in a complex way until it finally reaches the output layer and undergoes a thorough conversion. During the training process, the weights and thresholds will be adjusted continuously until the training data with the same label always produces similar outputs.
Deep learning is basically machine learning on steroids. There are multiple processing elements, and usually, each layer extracts some valuable information. For example, a neural network can process images used to drive self-driving cars. Each layer will handle different content, for example, the first layer may be to detect the edges on both sides of the road. Another layer may be detecting lane lines in the image, and another layer may be other cars.
That is a word packed into a lot of information. NVDIA's images provide extremely compact visualization:
regardless of these differences, these three factors will continue to have a profound impact on our world today, and one day, it may bring us into a robot society.
reference
https://becominghuman.ai/ai-machine-learning-deep-learning-explained-in-5-minutes-b88b6ee65846