The term algorithm has recently become very popular. It has changed from just using mathematicians to what most marketing teams use to push artificial intelligence solutions.
In my project, I realized that some startups just use the word algorithm without actually explaining how they use the algorithm to make the project "AI." I hope this article will help you understand the difference between the "AI" algorithm and other algorithms.
What is an algorithm?
Ok, let's start with a simple definition.
algorithm:The set of procedures or rules to follow in calculations or other problem-solving operations, especially computers.
Basically, the goal of an algorithm is to solve a specific problem, usually defined by someone as a series of steps.
For example, a recipe for making a cake-this is an algorithm.
In other words, the algorithm is a shortcut to help us send instructions to the computer. The algorithm simply tells the computer what to do next through the "and", "or" or "not" statements. Obviously, like most math-related things, it starts out very simple, but becomes infinitely complex when expanded. It is important to note that not all algorithms are related to AI or machine learning.
The algorithm provides guidance for almost any AI system you can think of.
Sounds good, but in fact there are problems with traditional algorithms. In fact, you must tell you to create a step-by-step process to achieve your goals. Some computer algorithms are not designed to follow only explicit program instructions, but are designed to allow the computer to learn by itself.
机器 学习
What is it?
机器 学习: A set of algorithms that enable software to update and "learn" previous results without programmer intervention. It is provided by structured data to complete tasks without programming how to operate.
Machine learning consists of a series of algorithms. Basically, AI (machine learning is a subset of AI) is designed to learn in the same way as children. Because of the data set, AI can find patterns and build assumptions based on these findings.
This is calledModel-based learningIt allows AI to make better decisions than humans because it can consider more factors and analyze them in milliseconds.
The algorithm is like following a recipe. You follow a series of instructions: Prepare the ingredients, heat the oven to 200c, and then bake 10 minutes. The output/result will be a great cake.
Now let's imagine your oven is too hot. Through machine learning, the system learned from the past that the oven became too hot, so shut it down.
Machine learning can be compared to experienced chefs (If you have a good data set). It knows recipes and has learned a lot from previous experiences. For example, the system has found that this ingredient works well with this cake (Based on data) and can make suggestions/forecasts.By using ML, you get something beyond the sum of its parts.
As we have seen, the algorithm is a mathematical technique. It is derived by statisticians and mathematicians for specific tasks, such as prediction.
The algorithms in machine learning are not new. At the most basic level, machine learning programs are code. Therefore, they are code written in Python or Java or in a programming language. However, the practicality of the algorithm increases only when they are implemented in code, because computers can handle high computations.
There are many algorithms for AI, but I think they usually have 3 classes:
You certainly have subclasses of this 3 class, and you can even find a hybrid approach that tries to contain the power of both.
The difference between AI and "smart" algorithms
The difference between AI and clever algorithms is how it is programmed.
As a user, we tend to focus only on the moments of input and input them into the system, where the output is produced as a result of the system.
However, the most important thing that happens in the middle is: Let's call it a hidden step.
The hidden steps are usually unknown, so it is difficult to distinguish between AI and algorithms.
Some elements help us classify the system using AI and clever algorithms
- Basic algorithm
If the defined input results in a defined output, the system's itinerary can be classified as an algorithm.This program mimics the basic computing power behind formula decisions.
- Complex algorithm
If a complex set of rules, calculations, or problem solving operations can result in a defined output, then the journey of the system can be classified as a complex algorithm.
- Machine learning (AI)
In an AI system, the output is not deterministic, but is specified based on a complex mapping of the data and then multiplied by each output.
Machine learning makes assumptions, re-evaluates models and re-evaluates data, all without human intervention. This is a game that changes the rules of the game. Basically, human engineers don't need to code for every possible action/response. The ML system will find all possible patterns at speeds and abilities that humans cannot achieve.
Traditional algorithm.
It requires some input and some logic in the form of code and provides you with output.
Traditional algorithms are based on the steps described in the algorithmproduceOutput. AlgorithmenterIt is based onHard code yourselfRules and参数generateOutput.
Machine learning algorithm.
It requires input and output and gives you some logic that can then be used to process new inputs to provide output. The logic generated is the reason for this ML.
The ML algorithm is based on learning through the input provided to it.predictionOutput. Learning through input is called a training process.
Give the algorithm to learndataAnd adjust参数To explain the data. You can then use these parameter sets to interpret/predict new data.
But not all ML algorithms are based on neural networks, and algorithms that solve many business use cases can passreturnOrTree-based algorithmTo solve.
It’s all about complexity...
These algorithms typically require less computation and less data to perform quite well on many issues that are perfect for certain business problems. When you seeUsed to define these algorithmsOfWhen the term AI is used, I believe it has no meaning.
Most AIs modify their algorithms in some way. In other words, the same input does not need to produce the same output/response later. For example, the neural network modifies the "weights" of certain joints in its path based on the previous guess/correctness of the response to the input.
Neural Networks:A series of algorithms that try to identify potential relationships in a set of data by mimicking the process of how the human brain works. When linked together, algorithms (such as lines of code) become more robust. They are combined to build an AI system like a neural network.
The behavior of a machine learning algorithm depends on what it learns during training, and then how it resembles in real life-in production.This is very different from most common algorithms, which require companies to evaluate model performance in a way that is unique to machine learning algorithms.
ML and classical algorithms
I have seen customers apply similar upgrades to machine learning and classic algorithm projects, where you have some experimental development (PoC) and then full production.
However, everything else is completely different. Let me give you a few examples.
- ML solution is true AI if it is not programmedTo perform tasks, but to be programmed to learn to perform tasks
- Traditional learning methods, such as training models based on historical training data and evaluating outcome models based on input data, are not feasible because the environment is always changing
- Traditional methods have more rigorous mathematical methods, andMachine learning algorithms are more data intensive
In most cases, one of the fundamental differences is that machine learning can have a range of results that are valid, but not necessarily determined in advance.
In my experience, machine learning still takes a lot of time to work perfectly. For example, an ML project running in production may not show any errors at all, getting completely good behavior from any type of metric (such as CPU utilization), but still producing erroneous predictions.
In summary, traditional algorithms take some input and some logic in the form of code and encourage output. In contrast, machine learning algorithms use inputs and outputs and give some logic that can then be used to process new inputs to give an output.The logic generated is what makes it ML.
The ML algorithm can learn from the data, while the classic algorithm specifies the exact rules to find the overall answer.
This article is reproduced in the medium,Original address
Comments