Even with the support of artificial intelligence frameworks such as TensorFlow or OpenAI, artificial intelligence still requires deep knowledge and understanding compared to mainstream web developers. If you have already established a working prototype, you may be the smartest person in this room. Congratulations, you became a member of the Senior Club. On Kaggle, you can even make a substantial profit by solving real-world projects. All in all, it works well, but is it enough for you to start a business? After all, you can't change the market mechanism. From a business perspective, artificial intelligence is just another implementation of existing problems. Customers don't care what kind of implementation they take, they only care about the results. This means you can't just enjoy it with artificial intelligence. After the honeymoon period, you must create value. In the long run, only the customer is the most important. Although your customers may not care about artificial intelligence, VCs are very concerned about this. The news media is also concerned. This is true in many industries. This difference in focus may create a dangerous reality distortion for startups, but you should not be fooled: unless you create universal multi-purpose artificial intelligence, there is no free lunch. Even if you are the darling of VC, you must go to the end for your customers. So let us also be a driver and see how we prepare for the future. "Mainstream artificial intelligence train" Artificial intelligence seems to be different from other big trends, such as blockchain, Internet of Things, and financial technology. Of course, its future is unpredictable. But almost all technologies are like this. The difference is that our value proposition as a person seems to be at risk – not just other industries. Our value as decision makers and creators is being reassessed. This has caused people's emotional reactions. We don't know how to position ourselves. The number of basic technologies is very limited, and most of them can be classified as "deep learning", which forms the basis of almost all applications: convolutional neural networks, long and short-term memory networks, automatic encoders, random forests, gradient enhancement techniques, and A few other applications. Artificial intelligence also provides many other methods, but these core mechanisms have recently achieved overwhelming success. Most researchers believe that advances in artificial intelligence technology will come from improvements in these technologies (rather than those that are fundamentally different). For the above reasons, we can call this "mainstream artificial intelligence research." Any real world solution consists of these core algorithms and non-artificial intelligence shapes to prepare and process data (eg data preparation, functional engineering, environmental modeling). Improvements in this part of artificial intelligence tend to make non-artificial intelligence parts redundant. This is the essence of artificial intelligence, and it is almost the definition of it - making the way to solve specific problems obsolete. However, this part of non-artificial intelligence is often a true source of profit for companies driven by artificial intelligence. This is their secret weapon. Every improvement in artificial intelligence makes this competitive advantage more likely to be open source and accessible to everyone. But it can also have disastrous consequences. As Frederic Jelinek once said: "Every time I fire a linguist, the performance of the speech recognizer will improve." Machine learning has basically introduced the next phase of layoffs: code is reduced to data. Almost all recognition techniques based on models, probabilities and rules were eliminated in 2010 by deep learning algorithms. Now with hundreds of lines of scripts (plus a fair amount of data) you can beat domain expertise, functional modeling, and thousands of lines of code. As mentioned above: This means that proprietary code is no longer a defensive asset on the tracks of mainstream artificial intelligence trains. Significant contributions are extremely rare. Real breakthroughs or new developments, even new combinations of basic components, are only possible with very limited researchers. As you might expect, this inner circle is much smaller (the number of developers must be less than 100). Why is this? Perhaps this is rooted in its core algorithm: backpropagation. Almost every neural network is trained in this way. The simplest form of backpropagation can be learned in the calculus course of the first semester, which is completely uncomplicated (but not at the elementary level). While this may seem simple – or perhaps for this reason – in a rich history of more than 50 years, only a few people can see the difficulty and question its main structure. If backpropagation is as visible as it is today, our achievements may be 10 years ahead of the current stage (except for computing power). Electrode sheets can be divided into different electrode sheets according to different standards, for example: self-adhesive electrode sheets, if according to the material can be divided into 1. PET self-adhesive electrode sheet 2 silicone self-adhesive electrode sheet, 3 silicone self-adhesive electrode sheet 4, others Button self-adhesive electrode sheet; silica gel electrode sheet can be divided into water-absorbing electrode sheet, heating electrode sheet, conductive electrode sheet and so on according to the purpose. In addition to the electrode pads, there are electrode wires and related physiotherapy products related to them. Conductive Sheet,Conductive Plastic Sheet,Thermal Conductive Sheet,Conductive Rubber Sheet Original Electronics Technology (Suzhou) Co., Ltd. , https://www.original-te.com