Shulou(Shulou.com)06/03 Report--

The era of great changes, the subversive influence of artificial intelligence AI, and the ten major trends of technology

Https://blog.csdn.net/weixin_42137700/article/details/85044414

Guide: we are in an era of great changes, there is no doubt that artificial intelligence (AI) has become one of the frontiers of science and technology, will have a subversive impact on many industries, and may reshape the company's talent strategy, operation model and cooperation model with customers in the future. Business leaders are prepared in advance to study how artificial intelligence will affect their business strategies in case they are left behind by the fourth industrial revolution.

1. Deep Learning: revealing the working principle of Neural Network

Summary: deep neural networks that mimic the human brain demonstrate their ability to "learn" from image, audio and text data.

Meaning: an accurate understanding of the working principle of deep learning will help to make it more developed and applied. For example, deep learning can provide reference for network design optimization and architecture selection.

2. Capsule network: simulating the visual processing advantages of the brain.

Summary: capsule network is a new type of deep neural network architecture, which can process visual information in the same way as the brain, which means that it can identify the logical and hierarchical relationship between features.

Significance: for typical image recognition tasks, the capsule network ensures a high accuracy of 50% by reducing the error, and the capsule network does not need so much training sample data. It is expected that capsule networks are widely used in many problem areas and deep neural network architectures.

3. Deep reinforcement Learning: interactive problem solving

Summary: deep reinforcement learning is a neural network algorithm that interacts with the environment through observation, action and reward. It has been used in game strategies, such as Atari and go, including the famous "AlphaGo", which defeated the human champion.

Significance: the acquisition of depth-enhanced learning ability is one of the important indicators for the commercialization of artificial intelligence applications. It can be trained through simulation without tagging data at all. More commercial applications that combine deep enhanced learning with agent-based (agent) simulation are expected to emerge in the coming year.

4. Generate confrontation network: network pairing promotes training and reduces processing burden.

Brief description: generating countermeasure network is an unsupervised deep learning system composed of two competing neural networks-"generating network" produces false data that looks like real data sets, and "judgment network" absorbs real and synthetic data.

Meaning: the generated countermeasure network further expands the deep learning, enables it to handle a wider range of unsupervised tasks, and reduces the load required by the deep neural network. It is expected to see more commercial applications, such as using generation counternetwork technology to do network detection, and so on.

5. Simplify and enhance data learning: solve the challenge of data tagging

Summary: the biggest challenge of machine learning (especially deep learning) is the need to use a lot of tagged data to train the system. There are two widely used techniques to help solve this problem: (1) synthesize new data (2) transfer the training model of one task or area to another, such as "transfer learning" skills (transferring lessons learned from one task / area to another), or "one-time learning" skills (extreme transfer learning, learning through only one example or no related examples), thus making them "data reduction" learning skills.

Meaning: using these techniques, we can solve more problems, especially when there is less historical data. Expect to see more variants of streamlined and enhanced data, as well as different types of learning skills applicable to a wider range of business issues.

6. Probabilistic programming: a convenient language for model development

Summary: probabilistic programming is a high-level programming language and modeling framework, which allows developers to easily design probabilistic models and solve them automatically. Probabilistic programming languages allow us to reuse model libraries, support interactive modeling and authentication, and provide the necessary abstraction layers to deduce general model groups more widely and effectively.

Meaning: scenarios suitable for the probabilistic programming framework include situations of uncertainty and incomplete information that are very common in the business world. In the future, we will see more widespread use of these languages and expect them to be used for deep learning as well.

7. Hybrid learning mode: combine the advantages of algorithms to solve uncertain problems.

Summary: different types of deep neural networks, such as generating countermeasure networks and deep reinforcement learning, show great prospects in their effects and wide applications combined with different types of data. The hybrid learning model combines these two methods and can make full use of the advantages of each method.

Meaning: the hybrid learning model expands the types of business problems to in-depth learning of uncertainty. This can help us to achieve better results, improve the interpretability of the model, and thus encourage wider application.

Automatic machine learning: models can be created without programming

Summary: the development of machine learning model is a time-consuming and expert-driven task, including data preparation, feature selection, model or reporting selection, training and debugging. Automatic machine learning is designed to automate this work using a variety of different statistical and deep learning algorithms.

Meaning: automatic machine learning is seen as part of the "democratization" of artificial intelligence tools that allow users to develop machine learning models without advanced programming skills. This will speed up the creation of models by data scientists. We will see more commercial automatic machine learning packages and the integration of automatic machine learning with a wider range of machine learning platforms.

9. Digital contracture: virtual replicas beyond industrial applications

Brief description: digital Li Sheng is a virtual model, which is used for detailed analysis and monitoring of physical or psychological systems. The concept of digital Li Sheng body originated in industry and is widely used to analyze and monitor such as wind farms or industrial systems. Now, by using agent-based modeling (computational models used to simulate the behavior and interaction of automatic agents) and system dynamics (computer-aided strategy analysis and design methods), digital contractures are widely used in non-physical objects and process control, such as predicting customer behavior.

Meaning: digital twins can help promote the development and wider application of the Internet of things (IoT), providing a method for predictive diagnosis and maintenance of IoT systems. Looking forward to the future, it is expected to see more digital Li Sheng body use in entity system and consumer choice modeling.

10. Explainable artificial intelligence: open the black box

Summary: at present, there are many machine learning algorithms in use, which can perceive, think and act in a variety of applications. However, many of these algorithms are considered "black boxes", and people know almost nothing about how they calculate the results. Interpretable artificial intelligence is intended to further develop machine learning skills to produce more interpretable models while maintaining the accuracy of artificial intelligence predictions.

Meaning: explainable, verifiable and transparent artificial intelligence is critical to building trust in technology, which promotes the wider adoption of machine learning techniques. We predict that before starting to adopt artificial intelligence on a large scale, companies may use explainable artificial intelligence as a requirement or best practice, while the government may make interpretable artificial intelligence a regulatory requirement in the future.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.