News classification
Contact us
- Add: No. 9, North Fourth Ring Road, Haidian District, Beijing. It mainly includes face recognition, living detection, ID card recognition, bank card recognition, business card recognition, license plate recognition, OCR recognition, and intelligent recognition technology.
- Tel: 13146317170 廖经理
- Fax:
- Email: 398017534@qq.com
2017 machine learning years
2017 machine learning years
It is hard to believe that there are so many things that have happened in the field of artificial intelligence and machine learning this year, and it is difficult to make a comprehensive summary of the system. In spite of this, I tried to make a summary, hoping to help you review the extent to which today's technology has developed.
1.Alpha Go Zero: the rise of the Creator
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
If I have to choose the main highlight of this year, that is AlphaGo Zero. This new method not only in some of the most promising direction of improvement (such as depth of reinforcement learning), and it was demonstrated that this model can in no case study paradigm shift data of the translator (think: This is the change of thought, in business, to what is not a large amount of data of innovators chance). We've also recently seen Alpha Go Zero being promoted to other games of chess.
2.GAN: don't be afraid, just GAN
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
A recent study (meta-study). The report found the system index related research papers on GAN error. Nevertheless, it is undeniable that GAN continues to play its unique role, especially when it comes to the application of image space (for example, progressive GAN, conditional GANS or CycleGans in pix2pix).
The 3. deep learning version of NLP: the beginning of commercialization
This year's deep learning is the world of NLP, especially in translation. NLP has made us feel that translation is becoming easier and easier. Salesforce provides an interesting non autoregressive method to deal with complete sentence translation. Perhaps more pioneering is the unsupervised approach, UPV, provided by Facebook. Deep learning also helps businesses to make their recommendation systems better and perfect. However, a recent paper has also raised questions about recent advances, such as how simple kNN is compared to Deep Learning. As with the GAN research, it is not surprising that the astonishing speed of artificial intelligence research can also lead to the loss of scientific rigour. Although many or most of the progress of AI is from deep learning, there are many other aspects of innovation in AI and ML, which should be noticed.
4. problems of theory: interpretability and tightness
0? Wx_fmt=png
In relation to some of the problems mentioned above, many people criticize the lack of rigor and interpretability of the theoretical basis of this method. Not long ago, Ali Rahimi (Ali Rahimi) described modern AI as "alchemy" in his NIPS 2017 conversation. Yann Lecun responded quickly to a debate that could not be solved quickly. It is worth noting that a lot of efforts have been seen on the basis of trying to promote deep learning this year. For example, researchers are trying to understand how the neural networks are deeply generalized. Tishby's information bottleneck theory has also been debated for a long time this year as a reasonable explanation for some deep learning attributes. Are celebrating this year occupation career has also been questioned Hinton fundamental issues such as using backpropagation. Pedro Dominguez (Pedro Domingos) and other well-known researchers quickly entered the rhythm and developed deep learning methods using different optimization techniques. The last fundamental change proposed by Hinton is the use of (capsule) capsules (see the original) as a substitute for the convolution network.
5. service business: the better development experience
If we look at the engineering related achievements of AI, then over the past year, Pytorch has begun to stir up the craze, and become the real challenge of Tensorflow, especially in the aspect of research. Tensorflow responds quickly by publishing dynamic networks in Tensorflow Fold. There are many other battles between the big players in the "battle of AI", the most intense of which is around the clouds. All the major suppliers have stepped up to increase their AI support in the cloud. Amazon has been showing up in their AWS, big innovation, such as their recent performance of Sagemaker to build and deploy ML models. In addition, it is worth mentioning that smaller players have poured into.Nvidia, and recently launched their GPU cloud, which is another interesting option for training deep learning mode. All of these battles will undoubtedly promote industrial upgrading in the future. In addition, the new ONNX neural network indicates that standardization is an important and necessary step for interoperability.
6. the future social problems to be solved
0? Wx_fmt=png
In 2017, the social problems of artificial intelligence were also continued (upgrades). Elon Musk (Elon Musk) continue to push us closer and closer to the killer AI's idea, make many people feel depressed. There is also a lot of discussion about how AI will affect work in the next few years. Finally, we see more focus on the interpretability and bias of the AI algorithm.
7. new battlefields: Machine Learning + traditional industry
In recent months, I have been working on artificial intelligence in medical and medical fields. I am pleased to see that the speed of innovation in "traditional" areas such as "health care" is rapidly increasing. AI and ML have been applied to medicine for many years, starting with the expert system and the Bias system in 60s and 70s. However, I often find myself quoted a few months ago. Some recent innovations proposed this year include the use of Deep RL, GAN, or automatic encoders to help patients diagnose. Much of the recent advances in artificial intelligence have also been focused on precision medicine (highly personalized medical diagnosis and treatment) and genomics.
1.Alpha Go Zero: the rise of the Creator
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
If I have to choose the main highlight of this year, that is AlphaGo Zero. This new method not only in some of the most promising direction of improvement (such as depth of reinforcement learning), and it was demonstrated that this model can in no case study paradigm shift data of the translator (think: This is the change of thought, in business, to what is not a large amount of data of innovators chance). We've also recently seen Alpha Go Zero being promoted to other games of chess.
2.GAN: don't be afraid, just GAN
640? Wx_fmt=png&wxfrom=5&wx_lazy=1
A recent study (meta-study). The report found the system index related research papers on GAN error. Nevertheless, it is undeniable that GAN continues to play its unique role, especially when it comes to the application of image space (for example, progressive GAN, conditional GANS or CycleGans in pix2pix).
The 3. deep learning version of NLP: the beginning of commercialization
This year's deep learning is the world of NLP, especially in translation. NLP has made us feel that translation is becoming easier and easier. Salesforce provides an interesting non autoregressive method to deal with complete sentence translation. Perhaps more pioneering is the unsupervised approach, UPV, provided by Facebook. Deep learning also helps businesses to make their recommendation systems better and perfect. However, a recent paper has also raised questions about recent advances, such as how simple kNN is compared to Deep Learning. As with the GAN research, it is not surprising that the astonishing speed of artificial intelligence research can also lead to the loss of scientific rigour. Although many or most of the progress of AI is from deep learning, there are many other aspects of innovation in AI and ML, which should be noticed.
4. problems of theory: interpretability and tightness
0? Wx_fmt=png
In relation to some of the problems mentioned above, many people criticize the lack of rigor and interpretability of the theoretical basis of this method. Not long ago, Ali Rahimi (Ali Rahimi) described modern AI as "alchemy" in his NIPS 2017 conversation. Yann Lecun responded quickly to a debate that could not be solved quickly. It is worth noting that a lot of efforts have been seen on the basis of trying to promote deep learning this year. For example, researchers are trying to understand how the neural networks are deeply generalized. Tishby's information bottleneck theory has also been debated for a long time this year as a reasonable explanation for some deep learning attributes. Are celebrating this year occupation career has also been questioned Hinton fundamental issues such as using backpropagation. Pedro Dominguez (Pedro Domingos) and other well-known researchers quickly entered the rhythm and developed deep learning methods using different optimization techniques. The last fundamental change proposed by Hinton is the use of (capsule) capsules (see the original) as a substitute for the convolution network.
5. service business: the better development experience
If we look at the engineering related achievements of AI, then over the past year, Pytorch has begun to stir up the craze, and become the real challenge of Tensorflow, especially in the aspect of research. Tensorflow responds quickly by publishing dynamic networks in Tensorflow Fold. There are many other battles between the big players in the "battle of AI", the most intense of which is around the clouds. All the major suppliers have stepped up to increase their AI support in the cloud. Amazon has been showing up in their AWS, big innovation, such as their recent performance of Sagemaker to build and deploy ML models. In addition, it is worth mentioning that smaller players have poured into.Nvidia, and recently launched their GPU cloud, which is another interesting option for training deep learning mode. All of these battles will undoubtedly promote industrial upgrading in the future. In addition, the new ONNX neural network indicates that standardization is an important and necessary step for interoperability.
6. the future social problems to be solved
0? Wx_fmt=png
In 2017, the social problems of artificial intelligence were also continued (upgrades). Elon Musk (Elon Musk) continue to push us closer and closer to the killer AI's idea, make many people feel depressed. There is also a lot of discussion about how AI will affect work in the next few years. Finally, we see more focus on the interpretability and bias of the AI algorithm.
7. new battlefields: Machine Learning + traditional industry
In recent months, I have been working on artificial intelligence in medical and medical fields. I am pleased to see that the speed of innovation in "traditional" areas such as "health care" is rapidly increasing. AI and ML have been applied to medicine for many years, starting with the expert system and the Bias system in 60s and 70s. However, I often find myself quoted a few months ago. Some recent innovations proposed this year include the use of Deep RL, GAN, or automatic encoders to help patients diagnose. Much of the recent advances in artificial intelligence have also been focused on precision medicine (highly personalized medical diagnosis and treatment) and genomics.