Deep learning is subject to large-scale hype. People can't wait to use neural networks in every place. But does it apply in every place? We will discuss it in the following chapters. After reading it, you will know the main drawbacks of neural networks, and when you choose the right type of algorithm for your current machine learning problems, you will have a rough guideline. You will also learn about the major issues in machine learning that we are now facing.
Why is deep learning motivated?Deep Learning has four main reasons for the ongoing hype: data, computing power, algorithms themselves, and marketing. We will discuss each of them in the following sections.
1. Data
One factor that has increased the popularity of deep learning is the large amount of data available in 2018 that has been collected over the past few years and decades. This allows neural networks to truly realize their potential because the more data they get, the better.
In contrast, traditional machine learning algorithms will certainly reach a level where more data will not improve its performance. The following chart illustrates this:
2. Computational power
Another very important reason is the computing power available now, which allows us to handle more data. According to Ray Kurzweil, a leader in artificial intelligence, computing power is multiplied by a constant factor (for example, doubled each year) in each time unit, rather than gradually increasing. This means exponential growth in computing power.
3. Algorithm
The third factor in improving the popularity of Deep Learning is the progress of the algorithm itself. Recent breakthroughs in algorithm development are mainly due to making them run faster than before, which makes it possible to use more and more data.
4. Marketing
Marketing may also be a very important factor. For decades, the neural network (first proposed in 1944) has undergone some hype, but in the past there was no time for anyone to believe and invest. The phrase “deep learning†gives it a new fancy name, which makes new speculation possible. This is why many people mistakenly believe that deep learning is a new creation field.
In addition, other factors have also contributed to the marketing of deep learning. For example, Sophia, the humanoid robot of Hanson Robotics, has caused widespread controversy among the general public and several breakthroughs in the major fields of machine learning, making it a mass media, etc. .
Neural network and traditional algorithmWhen you should use neural networks or traditional machine learning algorithms, this is a difficult question to answer because it depends to a large extent on the problem you are trying to solve. This is also due to "there is no free lunch theorem," which roughly shows that there is no "perfect" machine learning algorithm that can perform well on any issue. For each question, a specific method is appropriate and can achieve good results, while another method may fail, but this may be one of the most interesting parts of machine learning.
This is also why you need to be proficient in several algorithms, and why it is better to get good machine learning engineers or data scientists through practice. In this article will provide you with some guidelines to help you better understand what type of algorithm should be used.
The main advantage of a neural network is that it almost exceeds the capabilities of all other machine learning algorithms, but there are some drawbacks that we will discuss and focus on in this article. As I mentioned before, deciding whether or not to use deep learning depends on the problem you are trying to solve. For example, in cancer detection, high performance is critical because the better the performance, the more people can be treated. But also with machine learning problems, traditional algorithms provide more than just satisfactory results.
Black box
The most well-known shortcoming of neural networks may be their "black box" nature, which means that you don't know how and why neural networks will produce a certain amount of output. For example, when you put a cat's image into a neural network and predict that it is a car, it is difficult to understand what caused it to produce this prediction. When you have human-explainable characteristics, it is much easier to understand why it is wrong. In comparison, algorithms like decision trees are very easy to understand. This is important because in some areas interpretability is very important.
This is why many banks do not use neural networks to predict whether a person is creditworthy because they need to explain to customers why they did not obtain a loan. Otherwise, the person may feel that he has been wrongly threatened by the bank, because he does not understand why he did not get a loan, which may lead him to change his opinion of the bank. This is also true for websites like Quora. If they decide to delete a user account because of a machine learning algorithm, they need to explain to the user why they have completed it. I doubt whether they will be satisfied with the answers given by computers.
With the promotion of machine learning, other scenarios will be important business decisions. Can you imagine that the CEO of a large company will make a multi-million dollar decision without understanding why it should be done? Just because the "computer" said he needs to do this.
2. Development duration
Although libraries like Keras make neural network development very simple, sometimes you need more control over the details of the algorithm, for example, when you are trying to solve a problem in machine learning.
Then you may use Tensorflow, which provides you with more opportunities, but because it is also more complex, development takes longer (depending on what you want to build). Then for the company's management, if it's really worth their expensive engineers to spend a few weeks developing something, then the problem will arise, and the problem can be solved more quickly with simpler algorithms.
3. Data volume
Compared with traditional machine learning algorithms, neural networks usually require more data, at least thousands or even millions of tag samples. This is not an easy problem to solve. If other algorithms are used, many machine learning problems can be solved well with less data.
Although in some cases neural networks rarely process data, they do not handle it in most cases. In this case, a simple algorithm like Naive Bayes can handle a few data well.
4. Calculate expensive
In general, neural networks are computationally more expensive than traditional algorithms. The state-of-the-art deep learning algorithm, which enables successful training of truly deep neural networks, may take weeks to fully start training from scratch. Most traditional machine learning algorithms take less than a few minutes to several hours or days.
The computational power required for a neural network depends largely on the size of the data, but it also depends on the depth and complexity of the network. For example, a neural network with one layer and 50 neurons will be much faster than a random forest with 1000 trees. In contrast, a neural network with 50 layers will be much slower than a random forest with only 10 trees.
Now you may know that neural networks are more suitable for certain tasks, but not necessarily for others. You learned that large amounts of data, more computing power, better algorithms, and smart marketing have increased the popularity of deep learning and made it one of the hottest areas. Most importantly, you have learned that neural networks can almost defeat all other machine learning algorithms and their attendant disadvantages. The biggest drawbacks are their "black box" nature, increased development time (depending on your problem), the amount of data needed, and most of their computational costs.
in conclusionDeep learning may still be a bit overly hyped at present and it is more than expected. But this does not mean it is useless. I think we live in the revival of machine learning because it is becoming more and more democratized, and more and more people can use it to build useful products. Machine learning can solve many problems and I believe this will happen in the next few years.
One of the main issues is that only a few people understand what they can do with it, and they know how to build a successful data science team that brings real value to the company. On the one hand, we have doctoral engineers who are the theoretical genius behind machine learning but may lack a business understanding. On the other hand, we have executives and management positions who do not know what deep learning can do and believe that it will solve all problems in the coming years. We need more people to fill this gap. This will produce more products that are useful to our society.
China leading manufacturers and suppliers of Ev Charger,AC Charging Station, and we are specialized in AC Charging Pile,AC Ev Charger, etc.
Ev Charger,AC Charging Station,AC Charging Pile,AC Ev Charger
Shenzhen Hongjiali New Energy Co., Ltd. , https://www.hjlcharger.com